Run the three test suites in parallel

[laughingman7743]

WHAT

.github/workflows/test.yaml no longer chains test → test-sqla → test-sqla-async with needs:. The three suites start together. Each stays a separate job, so "Re-run failed jobs" still works per suite.

WHY

The chain came from 8c3be28 ("Split CI test jobs for independent retries"). It split one job into three so that failed jobs could be re-run separately; it was not there to limit load. The suites share no named AWS resources:

• tests/pyathena creates a random pyathena_test_<id> schema per process, with S3 data under it.
• The SQLAlchemy suites create <ident>, <ident>_test_schema and <ident>_test_schema_2 from the plugin's follower ident.
• Tests in the shared S3 Tables namespace use a random table-name suffix.

Measured on run 36023393076 (#803, before this change):

• Total: 46.5 min.
• test took 10–13 min, test-sqla 13–15 min, test-sqla-async 16–17 min, each stage 5 Python versions × 8 pytest workers.
• In parallel, the wall time should approach the longest suite plus setup.

Load: with the suites in parallel, about 120 pytest workers run at once instead of 40. The account's adjustable Athena quotas are:

• Active DDL queries: 100
• Active DML queries: 150

TooManyRequestsException from StartQueryExecution is retried by the default RetryConfig. Metadata throttling is answered from Glue since #803.

TEST

• just lint passed, and the workflow YAML parses.
• This PR's own CI run is the measurement. It will be recorded here: total and per-job durations compared with 46.5 min, failures, and any TooManyRequestsException or rerun seen in the logs.

🤖 Generated with Claude Code