Further optimize esbuild binary output

[samouri]

summary
As of the time of this writing, the closure compiled output is smaller than that created by esbuild.

	v0.mjs brotli	v0.mjs uncompressed
closure compiler on main	60.82	212.68
esbuild on main	63.25	232.76

There are four low hanging fruit that I believe can close the gap:

1. Closure is smarter about DCEing unused static functions from classes. esbuild does not do this due to a nuance with correctness. One simple solution is to rewrite our biggest offender, Services to be all toplevel exported functions instead of static class functions. This shaves <0.5kb br, 5kb uncompressed> from esbuild output.
2. DCE of polyfills: pjs-promises in particular is DCEd by closure but not by esbuild. This is another ~0.8kb br compressed.
3. Closure is very aggressive about inlining properties specified with @enum as opposed to forcing a property access. This is often smaller output as well. Unsure of size impact here.
4. The src/config code is in every single binary, but only needs to be in Runtime.

[samouri]

updates

• (1) Found a reasonable solution via runtime: convert services to standard object #36486
• (2) This is now complete via 🚮 Remove Promise polyfill #35838.
• (3) Exploring this via an FR to terser FR: inline values even when the whole object is required terser/terser#1098.
• (4) This is a bit tricky, because while deferring to AMP.config works for extensions, other items like ads still likely need the full configuration

--

With (1) and (2) alone the diff between CC and esbuild is now 1.7%

[samouri]

update

• (3) is now separately pursued via Lint enum usage so that they are fully DCE-able #36754

[samouri]

updates
Due to dropping the CE polyfill, all values have dropped by ~1.3kb 🎉 . This means that our "esbuild on main" branch is now smaller than Closure output was as of a month ago (60.66 vs. 60.93).

The table below shows current state, along with a few potential optimizations we could toggle on.

Each individually is an incredibly mild improvement and bears some risk.
Therefore I plan on test them out after the transition is complete.

	v0.mjs brotli	v0.mjs uncompressed
closure compiler on main	59.46	208.01
esbuild on main	60.66	218.51
esbuild w/unsafe	60.46	218.01
esbuild w/unsafe + pure_getters	60.21	217.16
esbuild w/unsafe + pure_getters + remove nth_identifier	60.12	217.13