perf: use serde_json_borrow for Synth output

[meskill]

Summary:
Replace using only async_graphql_value::ConstValue with serde_json_borrow::Value for the Sync output. Evaluation and store left intact and that uses ConstValue as before.

Issue Reference(s):
Fixes #... (Replace "..." with the issue number)

Build & Testing:

• I ran cargo test successfully.
• I have run ./lint.sh --mode=fix to fix all linting issues raised by ./lint.sh --mode=check.

Checklist:

• I have added relevant unit & integration tests.
• I have updated the documentation accordingly.
• I have performed a self-review of my code.
• PR follows the naming convention of <type>(<optional scope>): <title>

[meskill]

Wrk changes

posts title

before

Running 30s test @ http://localhost:8000/graphql
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     2.47ms   13.49ms 360.82ms   99.24%
    Req/Sec    17.39k     1.85k   22.15k    86.81%
  2075331 requests in 30.09s, 10.40GB read
Requests/sec:  68973.83
Transfer/sec:    354.02MB

serde_json_borrow

Running 30s test @ http://localhost:8000/graphql
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     2.50ms   14.78ms 397.01ms   99.22%
    Req/Sec    18.43k     2.19k   21.42k    89.15%
  2203809 requests in 29.23s, 11.05GB read
  Socket errors: connect 0, read 0, write 0, timeout 100
Requests/sec:  75406.44
Transfer/sec:    387.04MB

big query

before

Running 30s test @ http://localhost:8000/graphql
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     9.46ms    4.02ms 100.68ms   85.24%
    Req/Sec     2.70k   332.08     5.14k    86.24%
  322219 requests in 30.09s, 15.94GB read
Requests/sec:  10708.26
Transfer/sec:    542.35MB

serde_json_borrow

Running 30s test @ http://localhost:8000/graphql
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     9.07ms    3.22ms  90.35ms   77.74%
    Req/Sec     2.79k   323.80     3.60k    82.50%
  333021 requests in 30.08s, 16.47GB read
Requests/sec:  11072.28
Transfer/sec:    560.79MB

[meskill]

Flamegraphs

posts title

insert_key from 4.44% to 1.3%

big query

insert_key from 4.96% to 0.94%

[meskill]

The performance difference is noticeable, but for the big queries the impact is lower showing only around 3% improvement.

And this pr needs several more fixes to work because right now it's only possible to run the benchmarks and tests are failing:

• implement clone_from for case of object. The clone_from is introduced to be able to clone from one JsonLike to another JsonLike
• implement merge for instrospection query because we have different JsonLike entries now and it should be merged differently like it was with ConstValue
• fix the tests related to changes in exec signature