Improve cargo-zigbuild zig performance

[breezewish]

I recently discovered my C++ projects builds notably slower when using cargo-zigbuild, compared with a cmake toolchain I craft manually similar to https://github.com/mrexodia/zig-cross

After some debugging, I found that the initial process of fn execute_compiler (everything before actual zig Command::run) takes heavy amout of time, almost equal to the Command::run itself:

It would be great if the initial time could be reduced. Thanks!

[messense]

Maybe we can cache some of the repeated works.

[breezewish]

I guess most cost comes from spawning the Rust and Zig to checkout their versions. The versions could be checked when zigbuild is invoked and stored in zigcc-xx.sh.

[kassane]

output

$ zig cc -###                                                                          
clang version 20.1.2 (https://github.com/ziglang/zig-bootstrap daa53a0971085d5e5e4b20b9984b21182ecec266)
Target: x86_64-unknown-linux-musl
Thread model: posix
InstalledDir: /home/kassane/zig/0.15.0-dev.552+bc2f7c754/files

poop test:

$ poop -d 100 'zig cc -###' './.cargo/bin/cargo-zigbuild zig cc -- -###' 
Benchmark 1 (19 runs): zig cc -###
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          4.51ms ±  555us    4.01ms … 6.44ms          1 ( 5%)        0%
  peak_rss           41.7MB ± 70.6KB    41.5MB … 41.7MB          0 ( 0%)        0%
  cpu_cycles         3.44M  ±  298K     3.22M  … 4.60M           1 ( 5%)        0%
  instructions       3.80M  ± 2.71      3.80M  … 3.80M           1 ( 5%)        0%
  cache_references    262K  ± 3.18K      258K  …  272K           1 ( 5%)        0%
  cache_misses       84.8K  ± 1000      83.1K  … 86.4K           0 ( 0%)        0%
  branch_misses      28.2K  ± 1.08K     26.1K  … 30.5K           0 ( 0%)        0%
Benchmark 2 (3 runs): ./.cargo/bin/cargo-zigbuild zig cc -- -###
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          79.2ms ±  355us    78.8ms … 79.6ms          0 ( 0%)        💩+1657.4% ± 15.5%
  peak_rss           41.7MB ±    0      41.7MB … 41.7MB          0 ( 0%)          +  0.1% ±  0.2%
  cpu_cycles          185M  ±  358K      184M  …  185M           0 ( 0%)        💩+5272.5% ± 11.5%
  instructions        346M  ± 90.3K      346M  …  346M           0 ( 0%)        💩+8993.8% ±  1.0%
  cache_references   12.1M  ± 44.7K     12.1M  … 12.2M           0 ( 0%)        💩+4544.0% ±  7.2%
  cache_misses       1.96M  ± 2.77K     1.96M  … 1.97M           0 ( 0%)        💩+2217.5% ±  2.0%
  branch_misses      1.41M  ± 4.04K     1.41M  … 1.42M           0 ( 0%)        💩+4906.7% ±  7.5%