fix(prs): reuse GitHub data and defer optional reads - #9835

1. browse_code — apps/server/src/pullRequest/PullRequestService.ts

2. browse_code — apps/server/src/sourceControl/githubGraphQlBudget.ts

💭 githubGraphQlBudget.ts checks out as conventional. Now I'm reviewing PullRequestService.ts, checking whether Clock and Duration are properly imported as namespaces, then considering whether the web app files fall within scope.

3. browse_code — apps/server/src/pullRequest/PullRequestService.ts

4. browse_code — packages/client-runtime/src/state/pullRequests.ts

💭 The imports seem standard, so now I want to check pullRequests.ts and pullRequestList.logic.ts for named imports from "effect", even though they aren't Effect services themselves. Let me look at pullRequests.ts around lines 120-180.

5. browse_code — apps/web/src/state/pullRequests.ts

💭 The React state and pull request logic changes look fine—no Effect conventions violated. Checking the PullRequestService changes now, I'm looking at the listStats implementation using Effect.fn with a generator, which matches existing patterns, and examining the mutable recentStats Map to confirm it's properly scoped as closure state within make. Tests were added covering these changes, and I don't see any lint-disable directives or error handling changes in the diff. I notice a minor variable shadowing issue in the listStats loop but that's a code-quality concern rather than an Effect convention violation, so it's outside my scope here. Let me also verify there are no bare "effect" imports being used anywhere in the changed files.

6. browse_code — In apps/web/src/components/pullRequest/pullRequestList.logic.ts, list the import statements at the top of the file. Also confirm whether rankPullRequestsByMergeReadiness still has any remaining callers passing a hasMeasuredSize argument and whether hasMeasuredSize is still used/defined in sortPullRequestGroups.

💭 No convention violations found there. Checking test coverage for the backend changes, I see the githubGraphQlBudget and PullRequestService caching changes both have tests, and there's a new test covering the recoverTransientFailure logic change for detecting revision changes during diff reads. I'm also verifying the summary path removal and the Effect.fn typing on listStats look correct, and confirming Clock.currentTimeMillis is used properly instead of Date.now. Nothing here seems to violate the conventions I'm checking, so I'm leaning toward reporting this as clear.