docs: correct README layout + ROADMAP after llmgate extraction

README:
- Project layout now reflects the actual tree (pkg/* rather than
  internal/*, which only holds api/ today).
- Parser link points to pkg/parser/ (was internal/parser/, which never
  existed).
- Add a short note that LLM provider access lives in the standalone
  llmgate module.

ROADMAP:
- Phase 2 "Live LLM clients" consolidated: Anthropic / OpenAI / Gemini
  are all live via llmgate, not a bespoke HTTP client. Retries and
  cost tracking moved to llmgate middleware. CountTokens and streaming
  are tracked in the llmgate roadmap rather than duplicated here.
- Phase 2 now reads as "shipped, minus benchmarks" — matches reality.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: hallelx2

README.md

@@ -225,7 +225,7 @@ Or via environment variables: `VLE_TLS_CERT_FILE`, `VLE_TLS_KEY_FILE`.
 | PDF | `ledongthuc/pdf` | Font-size heuristic recovers headings from unstructured PDFs |
 | Text | stdlib | Single-section fallback |
 
-New parsers drop in behind a one-method `Parser` interface — see [`internal/parser/`](internal/parser/).
+New parsers drop in behind a one-method `Parser` interface — see [`pkg/parser/`](pkg/parser/).
 
 ## Features
 
@@ -264,17 +264,24 @@ Track progress in [GitHub Issues](https://github.com/hallelx2/vectorless-engine/
 ```
 cmd/engine/            # main binary entry point
 internal/
-  api/                 # chi HTTP router, v1 routes
+  api/                 # chi HTTP router, v1 routes (private to the binary)
+pkg/
   config/              # YAML + env config with validation
-  llm/                 # provider-agnostic LLM client (Anthropic/OpenAI/Gemini)
-  queue/               # Queue interface + QStash/River/Asynq drivers
+  db/                  # pgx pool, embedded migrations, CRUD helpers
+  ingest/              # parse → persist → summarize pipeline
+  parser/              # Parser interface + MD / HTML / DOCX / PDF / TXT drivers
+  queue/               # Queue interface + QStash / River / Asynq drivers
   retrieval/           # Strategy interface + single-pass / chunked-tree
   storage/             # Storage interface + local / S3 drivers
   tree/                # core tree / section data model
-pkg/                   # (reserved for public Go types — stable surface)
 docs/                  # API spec, architecture notes, images
 ```
 
+LLM provider access lives in a separate module, [`llmgate`](https://github.com/hallelx2/llmgate),
+which the engine imports as `github.com/hallelx2/llmgate`. That's
+where Anthropic / OpenAI / Gemini clients, retry / budget / cache
+middleware, and the cost table live.
+
 ## Contributing
 
 Contributions are very welcome — especially parsers, benchmarks, and new LLM / storage drivers. Please open an issue first for anything non-trivial so we can align on the design.

ROADMAP.md

@@ -79,16 +79,17 @@ One-line: raw bytes → queryable, persisted tree.
 
 ---
 
-## Phase 2 — retrieval *(in progress)*
+## Phase 2 — retrieval *(shipped, minus benchmarks)*
 
 One-line: turn `POST /v1/query` from a 501 into the feature the engine exists for.
 
-- [~] **Live LLM clients** (swap out the stubs)
-  - [x] Anthropic (messages API via direct HTTP — no SDK dep, prompt caching via `cache_control` when `EnablePromptCache` is on, exp-backoff retries on 429/5xx, `/v1/messages/count_tokens` for real counts with a `len/4` fallback)
-  - [ ] OpenAI (responses or chat completions, structured outputs for section-ID selection)
-  - [ ] Gemini (generateContent, long-context mode for whole-tree single-pass)
-  - [x] Real token counting for Anthropic (count_tokens endpoint)
-  - [x] Retry with exponential backoff + jitter on 429 / 5xx (Anthropic)
+- [x] **Live LLM clients** — extracted to [`llmgate`](https://github.com/hallelx2/llmgate)
+  - [x] Anthropic, OpenAI, Gemini all live via langchaingo under a shared `llmgate.Client`
+  - [x] Provider switching is pure config (`llm.driver: anthropic | openai | gemini`)
+  - [x] Retry with exponential backoff + jitter on 429 / 5xx (`llmgate/middleware/retry`)
+  - [x] Cost tracking per call (`Usage.CostUSD` populated from a static price table)
+  - [x] Error classification shared across providers (`llmgate.Classify`)
+  - [ ] Real `CountTokens` via provider endpoints — currently heuristic in llmgate; tracked in the llmgate roadmap, not a blocker here
   - [ ] Streaming responses (SSE) — deferred to Phase 4
 
 - [x] **Retrieval strategies**