perf(encoding): reserve memory before the row encoding.

[st1page]

after #8692 we can get the row encoding size before we really encode a row. So we can reserve the buffer before encoding.

optimizing target: Row::value_serialize https://github.com/singularity-data/risingwave/blob/main/src/common/src/row/mod.rs#L80

bench:

cd ./src/common
cargo bench --bench bench_row -- basic

[st1page]

c.c. @BugenZhao feel free to comment if you have any idea, thanks!

[fuyufjh]

If estimate_size() is most likely to be more than the actual length, then it would be perfect, but I guess currently estimate_size() gives a average approximating size, which means, it could be more or less than the actuall length.

Please make sure to cover all data types in this micro-benchmark.