replace usage of slow slist

[MartinNowak]

• directly reset symbols after writing them to an object file
  (it's very cheap when they are still in cache)
• external references are already reset right away
  in outfixlist
• no need to reset symbols that cannot be referenced
  (e.g. locals, to SCenum/SCstruct symbols)
• reset tls helper with rtlsym_reset

[MartinNowak]

About 10% faster when compiling multilib archives.

[WalterBright]

From the autotest failures, it appears that some symbols are not getting reset.

[MartinNowak]

From the autotest failures, it appears that some symbols are not getting reset.

Unfortunately it's a bigger issue. Some obj implementations (macho, mscoff) retain a pointer to the symbols and defer writing the symbol table until all sections are written.
I worked on directly writing the symbol table (and patching it later) but it's too much work, see MartinNowak@fc413b7.
Instead it seems like a better idea to reset the symbols in the obj implementation where we know exactly which symbols were used.

[MartinNowak]

Updated, I had overlooked an spurious symbol_reset from an earlier approach to fix this issue.

[MartinNowak]

This seems to be a bit slower than directly resetting the symbols (prolly b/c they are out of cache when the obj file is finished), but it's dead simple and compiles static libraries 15-20% faster than w/ the current slist_reset of all symbols.
I originally tried to rewrite those obj modules that store pointers to symbols to directly convert the symbols into the target format (MartinNowak@fc413b7), but that's a major change of mscoff and macho.

[MartinNowak]

From the autotest failures, it appears that some symbols are not getting reset.

Found the issue with OMF, forgot to add pubdef symbols to the reset list.

[MartinNowak]

Fixed the merge conflict.

[andralex]

worksforme

[andralex]

Auto-merge toggled on

[qchikara]

This introduces a regression:
https://issues.dlang.org/show_bug.cgi?id=15729

[CyberShadow]

This pull request fixed a regression:
https://issues.dlang.org/show_bug.cgi?id=12624

Good news for a change!