CI: Reduce MR test matrix to speed up pipelines

Problem

The bash-test jobs run on self-hosted shell runners with only 2 runners available. This creates a catastrophic bottleneck:

• 10 bash-test jobs (PG 9.6 through 18) all compete for 2 runners
• Queue times reach up to 19+ minutes for later jobs
• Total queue time: ~35 minutes for jobs that only take ~5 minutes each

Analysis

Job	Queue Time	Actual Duration
bash-test-9-6	0.3s	320s
bash-test-16	0.7s	297s
bash-test-10	297s	232s
bash-test-11	320s	296s
bash-test-12	529s	267s
bash-test-13	616s	283s
bash-test-14	795s	286s
bash-test-15	898s	268s
bash-test-17	1084s	259s
bash-test-18	1168s	237s

Solution

Reduce the test matrix for MR pipelines from 10 versions to 2 versions (PG 16 + PG 17):

• PG 16: Latest stable version
• PG 17: Latest development version

Full test matrix (all 10 versions) will only run on:

• Main branch commits
• Tag releases

Expected Impact

• ~80% reduction in queue time (~28 minutes saved per MR pipeline)
• Faster developer feedback
• Same test coverage on main branch

Labels

ci, performance

---

Acceptance Criteria

• MR pipelines run a reduced test matrix of 2 PostgreSQL versions (PG 16 + PG 17) instead of all 10 versions (PG 9.6 through 18)
• Main branch commits and tag releases continue to run the full 10-version test matrix
• Expected pipeline time reduction: ~80% less queue time (~28 minutes saved per MR pipeline)
• CI configuration clearly documents which conditions trigger full vs. reduced test matrix

Definition of Done

• .gitlab-ci.yml updated with conditional test matrix logic (rules for MR vs. main/tag)
• MR pipeline verified to run only PG 16 + PG 17 bash-test jobs
• Main branch pipeline verified to run all 10 PostgreSQL version jobs
• No test coverage regression on main branch (all versions still tested before release)
• Pipeline time improvements measured and documented