[GitHub client] retry search pages independently and adapt page size on RESOURCE_LIMITS_EXCEEDED

[StanislavKozachenko]

Makes SearchUsers survive the occasional GraphQL errors that large locations hit, instead of failing the whole location (and, with the current workflow, the rest of the run).

Key changes:

• github/github.go — retryCount is reset after each successfully processed page, so maxRetryCount now means 10 consecutive errors rather than 10 errors per location (a large location is ~200 pages at --consider 1000)
• github/github.go — when every error in a response is RESOURCE_LIMITS_EXCEEDED, the page size is halved (5 → 2 → 1) before retrying, so the retried query is actually cheaper instead of failing the same way again; other errors keep the current behaviour
• github/github.go — the 10s retry pause is now a package variable (retryDelay) so tests don't sleep
• github/github_test.go — new tests using a fake net.Wrapper (no network): page size shrinks only on RESOURCE_LIMITS_EXCEEDED, and more than 10 non-consecutive errors no longer abort the search (fails on master with Too many errors received. Quitting.)

Verified with go vet ./..., go build ./..., go test ./..., and real runs of the built binary (--preset belarus, --preset united_states, --output yaml), both completing with exit code 0 and unchanged output format. No GraphQL errors happened during those two runs, so the retry/page-size path itself is covered by the unit tests rather than the live runs.

Merge order: this only touches github/github.go (+ a new test file), so it's independent of #141 (workflow changes) and can be merged before or after it — together they address #143. It also complements #137/#138, which reduce the query cost upfront; this adapts when a page is still too expensive.

Closes #145
Refs #143

[StanislavKozachenko]

@ashkulz friendly ping — would appreciate a review when you have a moment. Thanks!