Performance improvements for beam-postgres

[LaurentRDC]

This PR is about improving performance of beam-postgres to match postgresql-simple.

Fixes #769

[LaurentRDC]

The baseline is:

$ BEAM_BENCH_ROWS=100000 cabal -v0 bench beam-backend-bench:beam-postgres
Starting benchmark with 100000 rows
(set BEAM_BENCH_ROWS to override)
All
  SELECT * FROM imdb_names (100000 rows)
    postgresql-simple: OK
      249  ms ±  12 ms
    beam-postgres:     OK
      321  ms ± 8.8 ms

This is showing a 50% slowdown in beam-postgres vs postgresql-simple

[LaurentRDC]

After removing some redundant work from parsing every row:

$ BEAM_BENCH_ROWS=100000 cabal -v0 bench beam-backend-bench:beam-postgres
Starting benchmark with 100000 rows
(set BEAM_BENCH_ROWS to override)
All
  SELECT * FROM imdb_names (100000 rows)
    postgresql-simple: OK
      243  ms ±  13 ms
    beam-postgres:     OK
      263  ms ± 7.0 ms

beam-postgres is now showing a ~10% slowdown over postgresql-simple

[LaurentRDC]

I believe that the remaining slowdown is due to the use of free monads. This means that speeding up the beam-postgres benchmark would require a breaking change in beam-core, such that backends would need to adjust.

Thankfully, this shouldn't be too bad, as there is only one "third-party" backend released on Hackage (beam-mysql) which would need quite a bit of work to bring back to life anyways.

In any case, this will be left for further work

[LaurentRDC]

I am perplexed. The benchmarks on my local machine are showing that this PR closes the gap between beam-postgres and postgresql-simple. However, Github Actions are NOT showing these gains. I can't explain this