[DO NOT MERGE] feat: reduce allocs & improve throughput

[jensneuse]

...by introducing a limit buffer pool and fetch buffer heuristics

I figures out that we can reduce allocs by almost 50% and increase throughput by 30% by doing 2 changes:

1. add a limit buffer pool to resolvable instead of using sync.Pool or no pool at all
2. add heuristics to loader to estimate the ideal size of fetch buffers

Both changes reduce bytes.growSlice operations to almost zero in hot paths, see below.

before:

after:

[jensneuse]

We want to merge this at a later stage once we have better testing architecture in place.