fix(translator): pad empty text for thinking-only Claude stops

[rome-xi]

Why

Antigravity→Claude and Gemini→Claude streams (and non-stream Messages) can end normally with only thinking and no text or tool_use. Claude Code SDK treats that silent end_turn as an incomplete turn, retries, and eventually reports a fake 32k output-token error.

A legal empty Claude text content block before message_delta / message_stop lets the SDK complete the turn.

What

• Add shared Claude SSE empty-text pad helpers.
• Inject that empty text block on thinking-only normal stops (end_turn / STOP / empty) in the Antigravity and Gemini Claude translators.
• Do not pad max_tokens / MAX_TOKENS, tool turns, or empty streams with no thinking.
• Do not forge max_tokens or change real truncation semantics.
• Native Claude passthrough is unchanged.

Tests

go test ./internal/translator/common/ -count=1 -run 'Pad|EmptyText'
go test ./internal/translator/antigravity/claude/ ./internal/translator/gemini/claude/ -count=1 -run 'ThinkingOnly|EmptyText|Pad'
go test ./internal/translator/common/ ./internal/translator/antigravity/claude/ ./internal/translator/gemini/claude/ -count=1

• Common helpers: pad on thinking-only end_turn / STOP / empty; no pad on max_tokens, existing text, tool_use, empty stream, or already padded.
• Antigravity stream + non-stream: thinking-only STOP emits empty text before message_delta; MAX_TOKENS does not pad.
• Gemini stream + non-stream: same.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-16T03:33:17.090037Z	b7dfa16	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b7dfa16dd6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Restrict padding to actual normal finish reasons

When a thinking-only response ends with an abnormal Gemini reason such as SAFETY, RECITATION, or MALFORMED_FUNCTION_CALL, this logic leaves stopReason as end_turn, so NeedsEmptyTextPad inserts a visible empty-text block and causes the client to accept the blocked response as a normal completion. Gate padding on the original finish reason being STOP/unspecified rather than the already-collapsed Claude stop reason; the non-stream and Antigravity padding paths have the same issue.

Useful? React with 👍 / 👎.