Unify Zig build targets

[xtexx]

Help wanted: I am lazy to impl.
Discussion needed: some design choices needed.

Rationale

Currently almost all (AFAIK, all) Zig packages (ghostty, ncdu, etc.) are compiled targeting either Zig baseline or build host. We should make them target the baseline of AOSC OS.

Background

• (hint: "Zig" refers to the programming language, "zig" refers to the implementation by Zig Software Foundation)
• By default, Zig build binaries targeting build host. [ref]
• Zig build-{exe,lib,obj} build target is specified with a target query (i.e. -target <arch><sub>-<os>-<abi>) and a CPU model string (-mcpu [cpu][feature set]).
• Zig build (Zig Build System) provides a "standard" (sic.) way to specify build target (i.e. -Dtarget=, -Dcpu). Packages may adopt their own way to specify target or disallow specifying build target (Very rarely; I haven't seen any packages doing so; These should be fixed in upstream).
• Some packages (e.g. ncdu) set build target in their own build script. We should override them.
• Note that some packages may want to cross-compile something intentionally. (AFAIK currently there isn't any case in AOSC OS)
• Note that AOSC OS may support more than one CPU feature sets on the same architecture (e.g. loongarch{,_nosimd}).

Proposal

• Sourcing target queries
  • A: Add ZIG_TARGET and ZIG_CPU variables to Autobuild.
  • B: Hard-code the AOSC baseline by patching zig; add special aosc_xxxx target strings (where xxxx is the Autobuild architecture name).
• Adopting target queries
  • A: Update all build scripts manually to use them.
  • B: When ABTARGET environment variable is set, let the standard functions of Zig Build System return the baseline target.
• Idiot-proofing, when AB environment variable is set
  • Reject all ZCU targeting non-AOSC-baseline targets, unless AOSC_ZIG_UNCHECK_TARGET environment variable is set.
  • Reject all ZCU with optimization modes other than ReleaseFast, unless AOSC_ZIG_UNCHECK_OPT_MODE environment variable is set.

[Cyanoxygen]

I would like to think about the possible (extra) work on maintaining the patches hardcoding zig to use the AOSC defined baseline. Anything else LGTM, especially when we are building Zig packages using Autobuild, which should control as many things as possible.