perf(ext/ffi): Optimize FFI Rust side type checks

[aapoalas]

Minor optimisation to FFI parameters' Rust side type checking / extraction.

Main change is avoiding ZeroCopyBuf in pointer extraction, as ZeroCopyBuf extracts and carries unnecessary information like the length of the buffer being pointed to. For FFI pointers, we are only interested in the actual pointer and not at all in the length of the buffer.

Additionally, double type checks are avoided by liberal application of if let Ok(value) = v8::Local::<v8::SomeType>::try_from(value). This doesn't show particularly high impact but maybe there's still something.

Relevant comparisons:

Bench	Pre-change (current main)	Post-change (this PR)
nop_u64()	447.64 ns/iter	450.94 ns/iter
nop_i64()	454.48 ns/iter	443.68 ns/iter
nop_usize()	448.33 ns/iter	449.93 ns/iter
nop_isize()	450.39 ns/iter	446.78 ns/iter
nop_buffer()	576.09 ns/iter	516.99 ns/iter
return_u64()	329.06 ns/iter	322.48 ns/iter
return_i64()	329.76 ns/iter	321.93 ns/iter
return_usize()	332.15 ns/iter	326.04 ns/iter
return_isize()	332.56 ns/iter	317.97 ns/iter
return_buffer()	331.82 ns/iter	320.1 ns/iter
nop_u64_nonblocking()	726.02 ns/iter	707.74 ns/iter
nop_i64_nonblocking()	713.36 ns/iter	712.2 ns/iter
nop_usize_nonblocking()	719.49 ns/iter	698.76 ns/iter
nop_isize_nonblocking()	709.2 ns/iter	707.5 ns/iter
nop_buffer_nonblocking()	851.82 ns/iter	773.95 ns/iter
return_u64_nonblocking()	532.23 ns/iter	545.31 ns/iter
return_i64_nonblocking()	534.77 ns/iter	528.94 ns/iter
return_usize_nonblocking()	539.35 ns/iter	541.26 ns/iter
return_isize_nonblocking()	517.91 ns/iter	530.17 ns/iter
return_buffer_nonblocking()	522.23 ns/iter	522.57 ns/iter
nop_many_parameters()	2.85 µs/iter	2.63 µs/iter
nop_many_parameters_nonblocking()	5.68 µs/iter	5.45 µs/iter

As can be seen, the BigInt types don't really change in a way that can be said to definitely have an effect. However, buffers are now measurably more performant.

[littledivy]

LGTM