perf: remove json_typeof

[steve-chavez]

Closes #2676. Removes json_typeof/jsonb_typeof by using some light validation on the json body using lazy bytestring functions.

• grants ~10% perf increase.
  • can be confirmed with the PlanSpec change.
  • can be confirmed manually with pgbench like:

$ postgrest-with-postgresql-16 -f test/pgbench/fixtures.sql pgbench -U postgres -n -T 10 -f test/pgbench/2676/old.sql 
tps = 3480.654744 (without initial connection time)

$ postgrest-with-postgresql-16 -f test/pgbench/fixtures.sql pgbench -U postgres -n -T 10 -f test/pgbench/2676/new.sql 
tps = 3973.138881 (without initial connection time)

• memory usage is kept the same
• no need for a new content-type as per a previous idea on Optimizations for INSERT #2676. This is an improvement for every write case.

[steve-chavez]

grants ~14% perf increase.

The above is true in pure SQL. But I haven't been able to reproduce a notable gain through http calls.

# Using https://github.com/PostgREST/postgrest-benchmark

$ postgrest-bench-k6 10 k6/POSTBulk.js # with 12.0.2

    http_reqs..................: 25959  822.396728/s

$ postgrest-bench-k6 10 k6/POSTBulk.js # with this PR

    http_reqs..................: 28556  906.419157/s

The loadtest in CI also shows the same.

[wolfgangwalther]

But I haven't been able to reproduce a notable gain through http calls.

822/s vs 906/s is about 10% more. What am I missing?

[steve-chavez]

Right, my bad. I was looking at some k6/POSTSingle.js (posts single object) results I had when I finished writing that. There isn't a noticeable gain there. But thinking more about it, it makes sense since json_typeof usage should mostly affect large payloads.