Split drop out of kafka codec benchmarks

[rukai]

Dropping types is actually quite expensive and affects benchmark results.
In this PR:

• To avoid measuring the drops of types we dont care about, we return them back to the criterion bencher as the return value.
• To measure the drops of types we do care about we create a separate benchmark just for measuring the time to drop.
  • We could keep them in the same benchmark which would reduce the number of benchmarks we have but splitting them like this makes it easier to tell the costs of different operations.

decode_all bench is deleted since we no longer perform batching from within decode so it no longer makes sense as a benchmark.

[github-actions]

0 benchmark regressed. 1 benchmark improved. Please check the benchmark workflow logs for full details: https://github.com/shotover/shotover-proxy/actions/runs/7839705427

Found 6 outliers among 100 measurements (6.00%)
  5 (5.00%) low severe
  1 (1.00%) low mild
kafka_codec/encode_request_metadata
                        time:   [388.34 ns 394.84 ns 401.15 ns]
                        change: [-40.401% -31.752% -25.337%] (p = 0.00 < 0.05)
                        Performance has improved.

[github-actions]

0 benchmark regressed. 2 benchmark improved. Please check the benchmark workflow logs for full details: https://github.com/shotover/shotover-proxy/actions/runs/7842754104

Found 7 outliers among 100 measurements (7.00%)
  4 (4.00%) low severe
  3 (3.00%) low mild
kafka_codec/encode_request_metadata
                        time:   [371.73 ns 378.05 ns 384.62 ns]
                        change: [-36.507% -31.704% -24.778%] (p = 0.00 < 0.05)
                        Performance has improved.
--
  1 (1.00%) low mild
  2 (2.00%) high mild
  5 (5.00%) high severe
kafka_codec/encode_request_list_offsets
                        time:   [530.83 ns 550.48 ns 573.16 ns]
                        change: [-36.879% -34.222% -31.322%] (p = 0.00 < 0.05)
                        Performance has improved.