turbopack: experimental minify-before-chunking feature
Minification happens on a per-chunk basis. With enough chunks, we get good
parallelization, but a large chunk can create a single large span that
wastes unused cores.
This flag makes minification more granular, applying it to each module and
caching the result. The goal is to make each minification step smaller, so
that they can be evenly distributed across parallel workers.
Theoretically this may also reduce the work done on a re-build, since the
minification transform will only be applied to changed modules.
The flag is configured behind an `experimental` config flag, and only
applies to builds (because we don't minify in dev!).
An ideal outcome is equal-or-faster builds, with almost no change to
byte size. I don't expect much change to byte size, becuase the shape
of our chunks doesn't lend itself to cross-module optimizations, but
it's entirely possible we regress this a bit. That hypothesis will
need to be tested against our benchmark suite of real Next apps.
