feat(llm): scale OpenRouter request timeout

[tom-sapletta-com]

Implements ticket 034 on the validated ticket-027 publication base.

The request deadline now scales deterministically at 1x, 2x, 4x and 8x from serialized input size, requested output tokens and structural complexity, with a 600 second cap. Retry backoff remains inside the same deadline and external cancellation remains immediate. Audit configuration records the non-secret policy, and timeout errors include base and effective durations.

Validation on the exact head:

• governance: PASS, 0 errors and 0 warnings
• npm run verify: PASS, 349 total, 348 passed, 1 optional JDK skip
• gold v2: PASS, all measured precision and recall 100 percent
• SDK examples: PASS across five languages
• Docker smoke: PASS

The local ignored OpenRouter configuration selects z-ai/glm-5.2 for the default and every stage. No live LLM call was made and no secret is part of this PR.

An independent protected GitHub review is required before merge.

[ifuri-validator-agent]

Deterministic Validator approval for exact head 68b0c0985f0aa95f8a41e252399491fe7aea29ca.

Ticket: ticket-034
Correlation ID: todo2code-pr-13-ticket-034-68b0c098
Model: openrouter/z-ai/glm-5.2
Advisory LLM verdict: REQUEST_CHANGES
Advisory summary: Adaptive timeout implementation is reasonable, but tests are truncated/incomplete and no test execution evidence is provided in the PR data.
Advisory findings: PR data is truncated mid-test-file and mid-test-case, making full review impossible. The final test case and file closing are missing.; No test execution results (passing CI logs) are visible; strict policy says never APPROVE without verified passing tests.
The LLM output above is advisory and was not used as the approval trust root.
Merge was not requested or performed.