fix(core): revert Linear Merge chunking — breaks merge semantics

Chunking Linear Merge causes each chunk to delete the previous chunk's
records (server thinks each subset is the full dataset). Reverted to
single-call approach.

The Antfly payload size limit (~6k docs) is an Antfly-side issue that
needs a fix in the server (raise JSON body limit or support streaming).
Tracked in scratchpad. chunk() utility kept — useful elsewhere.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: prosdev

.claude/scratchpad.md

@@ -4,7 +4,7 @@
 
 - **`getDocsByFilePath` fetches all docs client-side (capped at 5k).** Uses `getAll(limit: 5000)` + exact path filter. Fine for single repos (dev-agent has ~2,200 docs). Won't scale to monorepos with 50k+ files. Future fix: server-side path filter in Antfly SDK.
 - **Two clones of the same repo share one index.** Storage path is hashed from git remote URL (`prosdevlab/dev-agent` → `a1b2c3d4`). Two local clones on different branches share the same index, graph cache, and watcher snapshot. Stale data possible if branches diverge significantly. Pre-existing design — not introduced by graph cache. Fix would be to include branch or worktree path in the hash.
-- **Antfly Linear Merge fails on large batch sizes (~6k+ docs).** Tested with cli/cli (5,933 docs): `decoding request: json: string unexpected end of JSON input`. The scanner completes successfully but Antfly's HTTP endpoint fails to process the full payload. Workaround: none currently — the full index fails. Fix options: (1) batch the linearMerge into chunks of ~3k docs, (2) raise the limit on Antfly side, (3) stream instead of single POST. This blocks indexing medium-large Go/Rust repos (>~5k components).
+- **Antfly Linear Merge fails on large JSON payloads (~6k+ docs).** Tested with cli/cli (5,933 docs): `decoding request: json: string unexpected end of JSON input`. The scanner completes successfully but Antfly's HTTP endpoint can't parse the JSON body. Chunking is NOT a viable fix — Linear Merge semantics require ALL records in one call (the server deletes records not in the set, so each chunk deletes the previous chunk's data). Fix must be Antfly-side: raise the JSON body size limit, or support streaming/chunked transfer encoding. File a ticket with Antfly. Blocks indexing repos with >~5k components.
 - **Rust/Go callee extraction does not resolve target files.** tree-sitter callees have `name` and `line` but no `file` field (unlike ts-morph which resolves cross-file references). This means `dev_map` hot paths show 0 refs for Rust/Go repos, and `dev_refs --depends-on` won't trace cross-file paths. The dependency graph only has edges when callees include a `file` field. Future: cross-file resolution for tree-sitter languages.
 
 ## Open Questions

packages/core/src/vector/antfly-store.ts

@@ -7,7 +7,6 @@
  */
 
 import { AntflyClient } from '@antfly/sdk';
-import { chunk } from '../utils/chunking';
 import type {
   EmbeddingDocument,
   SearchOptions,
@@ -304,13 +303,6 @@ export class AntflyVectorStore implements VectorStore {
    * Use ONLY for full-index operations. For incremental updates, use batchUpsertAndDelete().
    * Records must be sorted lexicographically by key (handled internally).
    */
-  /**
-   * Maximum documents per Linear Merge HTTP request.
-   * Antfly's endpoint fails on large JSON payloads (~6k+ docs).
-   * Chunking into smaller batches avoids the limit.
-   */
-  private static readonly MERGE_BATCH_SIZE = 3000;
-
   async linearMerge(
     documents: EmbeddingDocument[],
     lastMergedId = '',
@@ -322,21 +314,51 @@ export class AntflyVectorStore implements VectorStore {
     this.assertReady();
 
     const sorted = [...documents].sort((a, b) => a.id.localeCompare(b.id));
+    const records: Record<string, unknown> = {};
+    for (const doc of sorted) {
+      records[doc.id] = { text: doc.text, metadata: JSON.stringify(doc.metadata) };
+    }
+
     const total = documents.length;
     const totals: LinearMergeResult = { upserted: 0, skipped: 0, deleted: 0 };
-
-    // Chunk documents to avoid Antfly HTTP payload size limit
-    const chunks = chunk(sorted, AntflyVectorStore.MERGE_BATCH_SIZE);
+    let cursor = lastMergedId;
 
     try {
-      for (const chunk of chunks) {
-        const result = await this.linearMergeChunk(chunk, lastMergedId);
-        totals.upserted += result.upserted;
-        totals.skipped += result.skipped;
-        totals.deleted += result.deleted;
-        if (result.took) totals.took = (totals.took ?? 0) + result.took;
+      const raw = this.client.getRawClient();
+      do {
+        const result = await raw.POST('/tables/{tableName}/merge', {
+          params: { path: { tableName: this.cfg.table } },
+          body: { records, last_merged_id: cursor },
+        });
+
+        if (result.error) {
+          throw new Error(
+            typeof result.error === 'object' && 'error' in result.error
+              ? String((result.error as Record<string, unknown>).error)
+              : String(result.error)
+          );
+        }
+
+        const data = result.data;
+        if (!data) {
+          throw new Error('Linear Merge returned no data');
+        }
+
+        totals.upserted += data.upserted ?? 0;
+        totals.skipped += data.skipped ?? 0;
+        totals.deleted += data.deleted ?? 0;
+        if (data.took) totals.took = (totals.took ?? 0) + data.took;
+
         onProgress?.(totals.upserted + totals.skipped, total);
-      }
+
+        if (data.status === 'partial' && data.next_cursor) {
+          cursor = data.next_cursor;
+        } else {
+          break;
+        }
+        // biome-ignore lint/correctness/noConstantCondition: pagination loop exits via break
+      } while (true);
+
       return totals;
     } catch (error) {
       throw new Error(
@@ -345,58 +367,6 @@ export class AntflyVectorStore implements VectorStore {
     }
   }
 
-  /**
-   * Merge a single chunk of documents via Antfly's merge endpoint.
-   * Handles server-side pagination (status: "partial" + next_cursor).
-   */
-  private async linearMergeChunk(
-    chunk: EmbeddingDocument[],
-    lastMergedId: string
-  ): Promise<LinearMergeResult> {
-    const records: Record<string, unknown> = {};
-    for (const doc of chunk) {
-      records[doc.id] = { text: doc.text, metadata: JSON.stringify(doc.metadata) };
-    }
-
-    const totals: LinearMergeResult = { upserted: 0, skipped: 0, deleted: 0 };
-    let cursor = lastMergedId;
-
-    const raw = this.client.getRawClient();
-    do {
-      const result = await raw.POST('/tables/{tableName}/merge', {
-        params: { path: { tableName: this.cfg.table } },
-        body: { records, last_merged_id: cursor },
-      });
-
-      if (result.error) {
-        throw new Error(
-          typeof result.error === 'object' && 'error' in result.error
-            ? String((result.error as Record<string, unknown>).error)
-            : String(result.error)
-        );
-      }
-
-      const data = result.data;
-      if (!data) {
-        throw new Error('Linear Merge returned no data');
-      }
-
-      totals.upserted += data.upserted ?? 0;
-      totals.skipped += data.skipped ?? 0;
-      totals.deleted += data.deleted ?? 0;
-      if (data.took) totals.took = (totals.took ?? 0) + data.took;
-
-      if (data.status === 'partial' && data.next_cursor) {
-        cursor = data.next_cursor;
-      } else {
-        break;
-      }
-      // biome-ignore lint/correctness/noConstantCondition: pagination loop exits via break
-    } while (true);
-
-    return totals;
-  }
-
   /**
    * Combined upsert + delete in a single batchOp call.
    * Safe for incremental updates and concurrent calls.