fix(sse): default summary + strip malformed id on kept Responses input items (#11108)

[maxmad64bis]

⚠️ base-red inherited: #9985

Summary

applyReasoningInputPolicy() / sanitizeResponsesInput() (open-sse/services/reasoningInputPolicy.ts) never guaranteed two fields stayed valid on input[] items it kept for replay:

1. summary on a kept type: "reasoning" item — never defaulted when the source omitted it entirely.
2. id on any kept item (reasoning and non-reasoning alike) — only ever stripped when it was already a valid string; a malformed id (observed as null, same upstream omission pattern) survived replay untouched.

If the upstream that produced the item never included summary (observed with opencode/muse-spark-1.2-contributor-free via the opencode/zen provider — its Responses API output omits summary entirely on opaque-only reasoning items, and separately sends id: null instead of a rs_... string on some items, including non-reasoning function_call items), that missing/malformed field survives the replay round trip. Some Responses-API-compatible upstreams treat both as required-if-present and reject the next turn:

[400] invalid_request_error: `input[N]` missing required field `summary`

Expected 'id' to be a string.

Same code path recently touched by #10949/#10959/#10961 (mixed plaintext/opaque reasoning replay) for the summary gap; the id gap is the sibling finding caught while validating that fix end-to-end against live muse-spark traffic — and the same class of gap was independently present in responsesInputSanitizer.ts::sanitizeInputItemId(), which returned a malformed id unchanged instead of stripping it.

Change

• sanitizeResponsesInput(), reasoning branch: default next.summary = [] on a kept type: "reasoning" item carrying opaque state when the source omitted it. Plaintext-only items intentionally have no summary key and stay untouched.
• sanitizeResponsesInput(), reasoning branch: keep id only when the item is opaque and id is a valid string; strip it otherwise (collapsed two overlapping conditions into one, via the new shared helper).
• sanitizeResponsesInput(), generic branch (function_call, message, ...): strip id whenever present, valid or not.
• New open-sse/services/responsesItemId.ts::isValidResponsesItemId(), shared by both places above and by responsesInputSanitizer.ts::sanitizeInputItemId(), which now actually strips a malformed id instead of passing it through unchanged.
• Flipped one legacy assertion in strip-reasoning-blobs-agentic-context-1599.test.ts that locked in the old byte-verbatim passthrough (no summary) to the corrected contract.

Test Plan

Regressions in tests/unit/reasoning-input-policy-summary-11108.test.ts and tests/unit/responses-input-sanitizer-name.test.ts:

• Compatible-transport passthrough defaults summary: [] on an opaque item missing it; an existing summary is left untouched.
• Incompatible-transport drop-and-repair path also defaults summary on the surviving opaque item.
• A non-string id (null) on a kept opaque reasoning item is stripped; a valid string id is preserved.
• A non-string id on a non-reasoning item (function_call) is stripped; call_id is untouched.
• responsesInputSanitizer.ts::sanitizeInputItemId() now strips a non-string reasoning-item id instead of passing it through (previously untested gap).

node --import tsx/esm --test tests/unit/reasoning-input-policy-summary-11108.test.ts tests/unit/responses-input-sanitizer-name.test.ts tests/unit/codex-system-message-image-8089.test.ts
# 26 pass, 0 fail

npx cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --import tsx/esm --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=1 \
  tests/unit/reasoning-cache.test.ts \
  tests/unit/strip-reasoning-blobs-agentic-context-1599.test.ts \
  tests/unit/translator-resp-openai-responses.test.ts \
  tests/unit/translator-openai-responses-req.test.ts \
  tests/unit/encrypted-reasoning-summary-7243.test.ts \
  tests/unit/reasoning-input-policy-summary-11108.test.ts \
  tests/unit/responses-input-sanitizer-name.test.ts \
  tests/unit/codex-system-message-image-8089.test.ts
# 152/152 pass, 0 fail (no regressions from the shared-helper refactor)

npx eslint open-sse/services/reasoningInputPolicy.ts open-sse/services/responsesInputSanitizer.ts open-sse/services/responsesItemId.ts tests/unit/reasoning-input-policy-summary-11108.test.ts tests/unit/responses-input-sanitizer-name.test.ts — clean.

npx tsc --noEmit -p tsconfig.json — no errors on the touched files.

Closes #11108