bug#53482: Byte-code speedup [PATCH]

[Mattias Engdegård]

The branch scratch/bytecode-speedup now contains a set of related improvements 
to speed up the byte-code interpreter. If someone is allergic to Git, I could 
post them as separate patches.

Performance is improved in many ways but is mostly focussed on making Elisp's 
lamentably slow function calls faster, which should benefit a lot of code. 
Operations on fixnums have also been sped up, as well as some data structure 
access.

Equally importantly, the C stack requirements has shrunk meaning that users can 
recurse deeper before getting a segfault. For example, bytecode is no longer 
copied to a stack buffer for execution. I intend to carry this process further 
in a set of follow-up patches, but the current changes are definitely good 
enough on their own.

I have run a wide set of benchmarks; regressions are rare but obviously not all 
code benefits equally. Most code becomes at least a little faster.

Unless there are good arguments not to, I'll merge it to master shortly.