Tracking issue for the LLVM 20 upgrade

[alexrp]

We've accumulated a number of issues that should be addressed with LLVM 20, so just opening this as a meta issue to track them all in one place:

• [WebAssembly] memcpy does not result in no-op for zero-length slices #16360
• behavior.vector.test.vector float operators causes LLVM assertion failure on mips*-linux-* #21051
• Remove "slow target" flag for mips(el)-linux targets in tests with LLVM 20 #21096
• Re-enable FastISel for MIPS O32 with LLVM 20 #21215
• Remove s390x-linux-gnu assembler workarounds with LLVM 20 #21329
• Remove the PowerPC soft float preprocessor workaround with LLVM 20 #21411
• Consider switching *-uefi-* targets to uefi instead of windows for LLVM 20+ #21630
• Update our baseline and generic CPU models for WebAssembly #21818
• Remove the LoongArch f16 lowering workaround with LLVM 20 #22003
• Re-enable native f16 lowering for Hexagon #22013

• zig/src/codegen/llvm.zig

  Lines 273 to 274 in a9c7714

  	.muslabin32 => "musl", // Should be muslabin32 in LLVM 20.
  	.muslabi64 => "musl", // Should be muslabi64 in LLVM 20.

• zig/src/target.zig

  Lines 466 to 474 in a9c7714

  	// This special-casing should be removed with LLVM 20.
  	switch (target.cpu.arch) {
  	.mips, .mipsel => return "o32",
  	.mips64, .mips64el => return switch (target.abi) {
  	.gnuabin32, .muslabin32 => "n32",
  	else => "n64",
  	},
  	else => {},
  	}

• zig/lib/std/Target.zig

  Lines 2266 to 2273 in a9c7714

  	.aarch64,
  	.aarch64_be,
  	.loongarch64, // TODO: `-sp` and `-sf` ABI support in LLVM 20.
  	.m68k,
  	.powerpc64,
  	.powerpc64le,
  	.s390x,
  	=> |arch| if (abi == .musl) initFmt("/lib/ld-musl-{s}.so.1", .{@tagName(arch)}) else none,