fix(tui): label session-scoped approval honestly, not "always" (#3766)

[Hmbown]

Summary

Closes #3766. Approval option 2 / a maps to ReviewDecision::ApprovedForSession (stored only in the in-memory app.approval_session_approved set), but the UI said "always" / "Approve always for this kind" — implying permanence. The actual cross-session persistence path is the separate [s] save-rule action. This corrects the copy so the trust boundary is honest.

Changes

• Compact controls (crates/tui/src/tui/widgets/mod.rs): [a] now renders session (en) / 本会话 (zh-Hans) instead of always / 始终.
• Full option label (ApprovalOptionApproveAlways): now session-scoped in en, es-419, ja, pt-BR, vi. zh-Hans / zh-Hant already said 本会话 / 本會話 and are unchanged.
• Keyboard hints (a, option 2) unchanged.
• Regression test approval_option_two_reads_as_session_scoped_not_always asserts both full and compact paths say "session" and never "always".

Validation

• cargo test -p codewhale-tui --bin codewhale-tui approval → 167 passed (incl. benign_a_two_approves_for_session and the new test).

Acceptance criteria

• Compact controls no longer render "always" for the session option
• Full option text no longer renders "Approve always"
• Label clearly communicates current-session scope
• Approval stays session-scoped (no accidental persistence)
• benign_a_two_approves_for_session still passes
• Added a copy regression test

[claude]

Claude finished @Hmbown's task in 1m 13s —— View job

---

PR Review: fix(tui): label session-scoped approval honestly, not "always"

• Read changed files and diff
• Verify locale changes across all locales
• Review widget code and regression test
• Summarize findings

---

Summary

This is a correct, well-scoped fix. Option 2 / [a] maps to ReviewDecision::ApprovedForSession, which is stored in the in-memory approval_session_approved set and does not persist across sessions. Calling it "always" was a trust-boundary lie. The fix is accurate.

What's Good

• The behavioral invariant is untouched — only the display string changes. benign_a_two_approves_for_session continues to pass.
• The regression test approval_option_two_reads_as_session_scoped_not_always covers both the full-card and compact paths and asserts the absence of "always" — exactly the right shape for a copy-correctness test.
• zh-Hans and zh-Hant already used session-scoped language (本会话/本會話), so those locales are consistent and correctly left unchanged.
• The comment added to push_compact_approval_controls explains the semantic link between [a] and ReviewDecision::ApprovedForSession, which is genuinely non-obvious to a future reader.

Findings

1. Stale identifiers now semantically wrong (not a blocker, but creates future confusion)

crates/tui/src/tui/widgets/mod.rs:1902 — the function is still called option_approve_always and dispatches through MessageId::ApprovalOptionApproveAlways. The JSON key is also "ApprovalOptionApproveAlways". These names now contradict the displayed text and the actual behaviour. This is all pre-existing naming, so it's out of scope to fix here, but it's worth a follow-up rename to option_approve_for_session / ApprovalOptionApproveForSession to stop future contributors from (incorrectly) concluding the option is persistent.

2. ZhHant (Traditional Chinese) compact controls still fall through to English (pre-existing, not a regression)

push_compact_approval_controls at line 1764 only special-cases Locale::ZhHans and falls back to English for everything else including ZhHant. Since zh-Hant.json already has a session-scoped full-card label (本會話同類自動批准), ZhHant compact controls showing "session" in English is inconsistent but also pre-existing. Not introduced by this PR.

3. Regression test uses only the default (en) locale

approval_option_two_reads_as_session_scoped_not_always checks "this session" and "always" which are English strings. It doesn't exercise zh-Hans, es-419, pt-BR, vi, or ja. That's acceptable coverage for a copy-correctness test (the JSON changes speak for themselves), but a quick assert!(!rendered.to_lowercase().contains("siempre")) for es-419 would be belt-and-suspenders. Optional, not required.

4. No correctness bugs, panics, or unsafe code introduced

All changed paths are pure rendering — no fallible operations, no new allocations beyond existing patterns.

Verdict

Ready to merge. The fix is correct and the regression test gives appropriate coverage. The stale option_approve_always / ApprovalOptionApproveAlways identifiers are a pre-existing naming issue worth a follow-up issue, not a blocker for this PR.