docs(pi): document image-strip omission in PARITY

The processed-image strip was already intentionally omitted on Pi (the
cleanup-stage notes in context-handler.ts explain the different image
part shape), but PARITY.md never recorded it, so parity audits kept
re-flagging it as an accidental gap.

Co-authored-by: Alfonso [Magic Context] <288211368+alfonso-magic-context@users.noreply.github.com>

Author: ualtinok

packages/pi-plugin/PARITY.md

@@ -557,6 +557,19 @@ channel) — it reflects `task_schedule_state` read-only.
 
 ---
 
+## 19b. Processed-image stripping is OpenCode-only
+
+OpenCode replaces large base64 image payloads in user `file` parts with
+sentinels once the assistant has processed them (`stripProcessedImages`,
+frozen-id replay). Pi's image part shape differs (`kind: "image"` with a URL)
+and large base64 images are rare in Pi sessions — the equivalent strip would
+only fire for the pasted-screenshot case, which Pi does not currently
+optimize. A Pi-specific image-content stripper can be added later if that
+case becomes common. Intentional divergence (see the cleanup-stage notes in
+`context-handler.ts`).
+
+---
+
 ## 19a. `/ctx-aug` skips empty sidekick augmentation blocks
 
 When sidekick returns the empty-result sentinel (for example, "No relevant