fix(llm): cap combined retry budget across transient and schema failures

[bh462007]

Fixes #735

What was wrong

llmCall() maintained two separate retry counters for each provider:

• retryCount for transient failures (max MAX_RETRIES = 2)
• schemaRetryCount for schema/JSON failures (max 1)

Although the loop checked both limits, each error path only considered its own counter before retrying. If a provider alternated between transient and schema failures, it could make 4 total attempts before falling back, exceeding the intended overall retry budget.

What changed

Added a totalAttempts counter with a limit of MAX_TOTAL_ATTEMPTS = MAX_RETRIES + 1 (3). Each retry now checks both its individual limit and the overall attempt limit.

This only affects mixed failure sequences. Existing behavior remains unchanged for:

• Transient-only failures (still up to 3 attempts)
• Schema-only failures (still up to 2 attempts)

I intentionally kept the existing retry counters instead of replacing them with a single shared counter, since that would have changed the schema-only retry behavior and broken an existing test.

Testing

• Added a test covering the reported sequence: transient → schema → transient → schema. It confirms retries now stop after 3 total attempts instead of 4.
• Verified that all 8 existing tests in router.test.ts continue to pass without changes, confirming no regressions for transient-only, schema-only, or timeout scenarios.
• Ran npx tsc --noEmit with no new TypeScript errors.

Note: I couldn't run the full test suite through the normal npm test path because tests/setup.ts is configured as a global setupFiles entry in vitest.config.ts and imports @testing-library/jest-dom/vitest, but the required Testing Library packages aren't listed in package.json. This prevents every test in the repository from running. To verify this change, I temporarily commented out that single config entry, ran the tests, then restored it before committing (git diff was clean afterward). I'm happy to open a separate issue for the test setup if that's helpful.

Scope

• src/lib/llm/router.ts
• src/lib/llm/router.test.ts

[vercel]

@bh462007 is attempting to deploy a commit to the codersogs-3057's projects Team on Vercel.

A member of the Team first needs to authorize it.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[pavsoss]

Hey @bh462007 the fix looks solid, but this branch contains unmerged commit from #761 , please rebase the branch for this PR on top of the main branch to isolate the LLM router change
Also it's recommended to first fetch changes from the upstream and then origin a new branch (for the new issues) on upstream/main

[jakharmonika364]

this branch have conflict with the main, please fix it

[bh462007]

Rebased my branch onto upstream/main using git rebase --onto upstream/main 7c4790a ee33b4a to remove the duplicate journey-progress commit, since it had already been merged upstream in #761 (5f36773). This left only the router fix in the branch.

I also added the upstream remote so I can base future branches directly on the latest upstream/main instead of my fork. Thanks for pointing that out!

[pavsoss]

LGTM