ci: benchmark every pull request against its base branch

[nigrosimone]

Runs an A/B benchmark on every pull request: the pg, pg-pool, pg-cursor and pg-query-stream of the PR against the same packages of the base branch, on the same runner in the same job, over the unix socket of the PostgreSQL that comes with the runner image.

Each arm runs in two processes, so every round also measures base against base and head against head. The same code on both sides gives the noise of that run on that machine, and a head/base ratio is marked only when it moved further than that noise, in every round.

The table goes to the job summary and to a comment on the PR. Only the ratio is comparable across runs, the absolute queries per second depend on the runner.

This is an example of the comment published on PRs

---

Benchmark: cbb85db against 9683053

Scenario	Base q/s	Head q/s	Head / base	Rounds	Noise
select 1 row, all types, prepared	7635	7760	1.025x (+2.5%)	1.00 to 1.03	±6.1%
select 100 rows, all types, prepared	1139	1138	0.997x (-0.3%)	0.99 to 1.01	±5.7%
select 500 rows, all types, prepared	278	276	0.987x (-1.3%)	0.98 to 1.00	±3.0%
select 500 rows, simple types, prepared	1602	1596	0.993x (-0.7%)	0.99 to 1.02	±3.8%
select 500 rows, simple types, array mode	1481	1495	1.012x (+1.2%)	0.99 to 1.02	±1.9%
select 500 rows, simple types, binary	826	818	0.989x (-1.1%)	0.99 to 1.02	±2.7%
select by id, parameterized	5518	5457	1.006x (+0.6%)	0.94 to 1.04	±3.9%
insert, all types, returning	4214	4233	1.005x (+0.5%)	0.96 to 1.04	±5.3%
update, parameterized	6179	6069	0.974x (-2.6%)	0.95 to 1.01	±5.3%
delete, parameterized	7214	7251	0.990x (-1.0%)	0.97 to 1.05	±3.4%
transaction, 3 queries	11974	12210	1.025x (+2.5%)	0.92 to 1.05	±7.1%
pool query, select 1 row	6938	7032	1.006x (+0.6%)	0.99 to 1.03	±7.2%
cursor, 500 rows in 5 reads	211	211	1.000x (+0.0%)	0.99 to 1.04	±3.8%
query stream, 500 rows	213	213	0.995x (-0.5%)	0.99 to 1.01	±1.8%
select 100 rows, all types, pipelined x10	1175	1193	1.010x (+1.0%)	0.99 to 1.02	±2.8%

4 rounds of 3s per scenario, each round measuring base, head and a second process of each one after the other, in an order that changes every round. "Head / base" is the median of the per-round ratios and "Rounds" their range. "Noise" is how far base/base and head/head, the same code on both sides, got from 1 in this same run: that is what the machine did, so a row is marked only when the median moved further than that, at least 2%, and every round moved the same way: 👀 slower, 🏆 faster. Only the ratio is comparable across runs, the absolute q/s depend on the runner.

Node v26.8.2, AMD EPYC 7763 64-Core Processor, 4 cores.