core: avoid async op future reboxing to bundle PromiseId

[AaronO]

This is another op-layer optimization, it reduces the baseline overhead of async ops by ~10%.

It removes an extra alloc/free per opcall that happened due to "reboxing" the future to insert the PromiseId. The PromiseId is now passed as a field of OpPayload so the async OpFn wrappers can insert it when mapping the future result.

I have another experiment in progress, but wanted to land this since it's simple and 10% is a respectable gain even if it's not "game changing".

Benches

Before:
test bench_op_async   ... bench:     533,080 ns/iter (+/- 14,308)
test bench_op_nop     ... bench:      71,169 ns/iter (+/- 2,998)
test bench_op_pi_bin  ... bench:      87,043 ns/iter (+/- 3,089)
test bench_op_pi_json ... bench:      85,530 ns/iter (+/- 1,801)

After:
test bench_op_async   ... bench:     474,119 ns/iter (+/- 8,209)
test bench_op_nop     ... bench:      71,417 ns/iter (+/- 3,776)
test bench_op_pi_bin  ... bench:      86,535 ns/iter (+/- 3,084)
test bench_op_pi_json ... bench:      86,308 ns/iter (+/- 1,388)

[piscisaureus]

Awesome!