test: speed up compose lifecycle

[sbalabanov]

Summary

• run test MySQL data directories on fresh tmpfs mounts and skip timezone-table initialization
• poll MySQL health every second while retaining a 60-second retry window
• retain service images under stable per-worktree names so up --build can reuse them without cross-worktree contamination
• keep database isolation by continuing to remove each unique Compose project's volumes with down -v

Benchmarks

Two isolated uncached Bazel runs per target on the same host:

Target	Before	After	Change
//test/integration/extension/counter/mysql:go_default_test	13.959s	10.096s	-27.7%
//test/integration/submitqueue/gateway:go_default_test	18.675s	12.904s	-30.9%
//test/e2e/stovepipe:go_default_test	15.909s	12.271s	-22.9%

Raw samples:

• counter: 13.874s, 14.043s -> 10.141s, 10.052s
• gateway: 20.860s, 16.489s -> 12.709s, 13.099s
• Stovepipe e2e: 15.811s, 16.007s -> 12.738s, 11.803s

Validation

• all 10 Docker-backed integration/e2e Bazel targets passed together
• make lint
• make check-tidy
• make check-gazelle

Tradeoffs

• test databases consume RAM while running; they remain fresh per stack and are discarded at teardown
• one finite set of service images is retained per worktree; --build still checks and refreshes them for correctness
• local/manual Compose retains volume-backed MySQL by default; the tmpfs setting is applied only by ComposeStack

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}