test(integration): add GetChangedTargets benchmarks

[yushan8]

Summary

• Adds a make bench target that measures GetChangedTargets latency with two benchmarks:
  • Cold — 15 iterations over unique commit pairs, each a cache miss exercising the full pipeline (git checkout, bazel query, compare, stream)
  • Cached — 15 iterations hitting the in-memory cache, measuring streaming overhead only
• Adds startServerWithLogger to allow benchmarks to inject a nop logger, suppressing service log noise
• Measurement only — no pass/fail thresholds, not part of make test / make test-integration, and never invoked from CI
• Widens repoRemote/startServer/newClient/getChangedTargets in integration_test.go from *testing.T to testing.TB so the existing integration harness can be reused from Benchmark* functions

Test plan

• make gazelle — integration/BUILD.bazel picked up the new benchmark_test.go
• make test-integration — existing integration tests still pass unchanged
• make bench — both benchmarks run 15 iterations and report ns/op, exits 0

[sywhang]

What's the current measured benchmark on this on your local box?

[yushan8]

What's the current measured benchmark on this on your local box?

BenchmarkGetChangedTargets_SmallDiff-10       3   503320 ns/op   0.0005030 avg_cached_call_sec/op   44.90 first_uncached_call_sec
BenchmarkGetChangedTargets_MediumDiff-10      3   466181 ns/op   0.0004658 avg_cached_call_sec/op   41.07 first_uncached_call_sec
BenchmarkGetChangedTargets_LargeDiff-10       3  1451639 ns/op   0.001451 avg_cached_call_sec/op    38.21 first_uncached_call_sec

uncached run is the initial request, cached run is the subsequent calls with the same request.

[sywhang]

let's run it with higher iteration by default. like 15 or 20.

[yushan8]

I decided changed the test to run getChangedTargets on different commits to test cold cache calls. Since each iteration needs a unique commit pair for a cache miss, b.N is capped at the number of pairs we have. BenchmarkGetChangedTargets_Cold tests for the requests with cache miss (full bazel query) with 15 distinctive commit pairs, while the other BenchmarkGetChangedTargets_Cached tests for the duration of cache hits.
This lets us measure the endpoint latency without the server startup.

[yushan8]

Updated benchtime to run 15x