chore(ci): batch unit tests, add timeouts, optimize Dockerfile

[pyramation]

Summary

Three optimizations to reduce CI costs and prevent waste:

1. Batch unit tests (22 jobs → 4 jobs)

Unit tests previously ran as 22 individual 2vcpu jobs, each with ~1 min of download/install overhead for tests that run in seconds. Due to per-minute billing rounding, this was billed as 44 2vcpu-minutes for ~5 min of actual test work.

Now grouped into 4 batched jobs:

Batch	Packages
uploads	mime-bytes, uuid-hash, uuid-stream, etag-hash, etag-stream, stream-to-etag, content-type-stream, upload-names
packages-core	url-domains, query-builder, csrf, oauth, 12factor-env, orm
packages-services	postmaster, smtppostmaster, csv-to-pg, cli, pgsql-client, pg-ast
graphql	query, codegen

Each batch runs tests sequentially using ::group:: markers for clear log separation. Estimated savings: ~73% of unit test minutes (44 → ~12 2vcpu-minutes).

2. Add timeout-minutes to all jobs

Prevents runaway jobs from burning money (default GitHub timeout is 6 hours):

• build: 10 min (usually ~2 min)
• unit-tests: 10 min (usually ~2 min)
• pg-tests: 15 min (usually ~5-8 min)
• integration-tests: 15 min (usually ~5-8 min)
• examples-types: 15 min

3. Dockerfile: pnpm fetch for layer caching

Adds a pnpm fetch layer that pre-populates the pnpm store using only pnpm-lock.yaml. This layer is cached and only invalidated when dependencies change — not when source code changes. Subsequent pnpm install --frozen-lockfile --offline links from the local store without network access.

Review & Testing Checklist for Human

• Verify all 4 unit test batches pass — confirm no test relies on being in its own isolated runner (e.g., global state pollution between tests in different packages within the same batch)
• Verify Dockerfile builds correctly — the pnpm fetch + --offline pattern should work with pnpm 10, but confirm the Docker image builds and the CLI shims work
• Spot-check timeout values — if any job routinely takes longer than the set timeout, it will start failing. The values are set at ~3× typical runtime but edge cases may exist

Notes

• pg-tests and integration-tests are left as individual matrix jobs since each has different env vars and test durations. Batching these would require more careful grouping.
• The ::group:: / ::endgroup:: markers in the batch test step provide collapsible sections in the GitHub Actions log so you can still see which specific test in a batch failed.

Link to Devin session: https://app.devin.ai/sessions/f42f42528fb0465684159227efbef017
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring