avoid allocation of closures

[rainers]

While debugging #23557 I noticed quite a few closure allocations happen during compilation. I guess these are unexpected in dmd, so let's get rid of them.

The assert in BumpGC.malloc triggers for such an allocation in debug builds, maybe it needs to be versioned for older host compiler versions or other backends that don't yet fully support the templated new.

[github-actions]

DMD perf check

Metric	Base	PR	delta
compile hello.d (instr)	216.1 M	215.6 M	-0.22%
compile hello.d -O (instr)	234.6 M	234.1 M	-0.21%
compile Phobos (instr)	5,137.2 M	5,110.8 M	-0.51%
compile Phobos codegen (instr)	1,474.6 M	1,472.4 M	-0.15%
compile vibe.d (instr)	15,161.3 M	15,081.3 M	-0.53%
dmd binary size (stripped)	6.92 MB	6.95 MB	+0.34%
hello binary size	0.72 MB	0.72 MB	0.00%
peak RSS (compile hello.d)	44 MB	43 MB	-0.24%
peak RSS (compile Phobos)	639 MB	618 MB	-3.24%
peak RSS (compile vibe.d)	1980 MB	1916 MB	-3.28%

[rainers]

Unfortunately, inferring scope for local functions regressed somewhere around dmd 2.101. I happened to have built with dmd 2.095, current versions need a lot more patches. Probably the regression should be fixed first.

[rainers]

Unfortunately, inferring scope for local functions regressed somewhere around dmd 2.101. I happened to have built with dmd 2.095, current versions need a lot more patches. Probably the regression should be fixed first.

Ah, my breakpoint in the BumpGC didn't fire because it is not built with the old versions. So not a regression...

[rainers]

Unfortunately, the diff here fails for hdrgen.d. It's mostly whitespace changes as functions had to be moved into the local class in toCBuffer, plus ref parameters buf and hgs have to be converted to pointers to capture them explicitly in the class.

[thewilsonator]

you should be able to keep the indentation level the same for most of the diff if you move this class outside of toCBuffer and move its definition to below the class