fix: close post-merge test gaps in hooks + Phase-2 history

[saucam]

What

Post-merge testing audit of #137 (hook bus) and #138 (Phase-2 history). The merged combination passes the full suite (1233 tests) and the new files measure 99–100% line coverage, but four behaviors were untested. This PR covers them — tests only, no product code changes.

New coverage

canonical-sdk-compat.test.ts — pins the riskiest converter integration points to the actual installed SDKs instead of code comments:

• toGeminiContent output is fed through the real @google/generative-ai startChat(), whose constructor runs validateChatHistory synchronously (offline, zero network). A negative control (functionResponse on a user turn — the design doc's original sketch) proves the validator actually fires, so the role-"function" decision can't silently regress on an SDK bump.
• toOpenAIMessages output is assigned to ChatCompletionMessageParam[] without a cast, so bun run typecheck enforces structural compatibility — the provider call site uses a cast, which would hide drift.

session-hooks.test.ts H8–H11:

• H8 — hook input mutation composed with manual approval (previous coverage was autonomous-mode only): the approval UI delta shows the hook-rewritten input, and the approved merge base is the mutated input, not the model's original.
• H9/H10/H11 — the provider_switched, rotated, and session_end observe emits, none of which were exercised before (only session_start/after_turn were).

Test-harness fix

The observe-hook assertions now wait for parseable JSON, not file existence — cat > file creates the file before the payload bytes land, which made H9/H11 flaky. H8 also waits for the mutation broadcast rather than waiting_approval (the tool_start handler flips that status from its side-effect-free peek before the hook gate runs, so status alone races the hook).

bun run test (1233 pass), typecheck, lint green.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.07%. Comparing base (8e9e86a) to head (d04cead).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #140   +/-   ##
=======================================
  Coverage   82.07%   82.07%           
=======================================
  Files          97       97           
  Lines       16867    16867           
=======================================
  Hits        13843    13843           
  Misses       3024     3024           

Flag	Coverage Δ
daemon	82.07% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[saucam]

/gemini-review