perf: enable "fat" LTO for production release

[chenjiahan]

Summary

Pros

• Better performance (according to benchmark): 1% ~ 3% faster
• Smaller binary size (rspack.darwin-arm64.node): 55.4 MB -> 51.7 MB

Cons

• Longer build time, only affect the production release: 4-5min -> 8-10min

see:

• perf: enable "fat" LTO for release #5507
• https://doc.rust-lang.org/cargo/reference/profiles.html#lto

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).

[netlify]

✅ Deploy Preview for rspack canceled.

Name	Link
🔨 Latest commit	8c8e09a
🔍 Latest deploy log	https://app.netlify.com/sites/rspack/deploys/668cda58ec84d10008d9c2d9

[chenjiahan]

!bench

[rspack-bot]

📝 Benchmark detail: Open

Name	Base (2024-07-09 f2bc023)	Current	Change
10000_development-mode + exec	2.18 s ± 17 ms	2.13 s ± 27 ms	-2.03 %
10000_development-mode_hmr + exec	688 ms ± 7.5 ms	693 ms ± 10 ms	+0.62 %
10000_production-mode + exec	2.81 s ± 26 ms	2.7 s ± 18 ms	-3.87 %
arco-pro_development-mode + exec	1.91 s ± 55 ms	1.89 s ± 68 ms	-0.83 %
arco-pro_development-mode_hmr + exec	433 ms ± 1.8 ms	433 ms ± 0.62 ms	-0.19 %
arco-pro_production-mode + exec	3.46 s ± 90 ms	3.43 s ± 56 ms	-0.94 %
threejs_development-mode_10x + exec	1.57 s ± 16 ms	1.58 s ± 12 ms	+0.57 %
threejs_development-mode_10x_hmr + exec	795 ms ± 8.4 ms	820 ms ± 3 ms	+3.23 %
threejs_production-mode_10x + exec	5.58 s ± 22 ms	5.55 s ± 55 ms	-0.46 %

[JSerFeng]

Cool, I thought we were already use fat in production release long ago

[chenjiahan]

I thought we were already use fat in production release long ago

Haha, I guess the reason for using "thin" before was mainly due to CI time. And it is worth to enable "fat" if it only affects the prod release