jsg: skip the fast-callback preamble and restore FastOneByteString

[littledivy]

I was digging into the JSG fast API implementation and found some overhead to remove. Skimming through the git history, it might be enough to make enabling fast calls worth another look.

This patch includes two changes:

1. Skip the HandleScope. Every fast callback opens a HandleScope, grabs the current context, and does two isolate lookups. That is all only needed to unwrap v8::Local arguments. If a method takes nothing but primitives and strings then there is nothing to unwrap, so none of it is needed.
2. Bring back FastOneByteString. It was removed in remove v8::FastOneByteString support #4628 because its data pointer dies on the next GC and unwrapping another argument could trigger one first. For these signatures no argument calls into V8 at all, so nothing can GC before we copy the bytes.

Ran //src/workerd/jsg:bench-dispatch with --config=benchmark and 3 reps:

	before	after
NativeDispatchFast	58.6 us	34.9 us
NativeDispatchFastArg	58.3 us	35.8 us

I also measured Headers locally:

method	slow	fast
has	52.5 ns	37.2 ns	-29.1%
set	101.3 ns	89.0 ns	-12.1%
delete	51.4 ns	43.2 ns	-16.0%

I will open follow up PRs that apply the same optimizations to the preamble in the 10 other fastCallback definitions.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[littledivy]

I have read the CLA Document and I hereby sign the CLA