Fix cache thrashing with runtime manganis support + introduce cross-platform asset resolver

[jkelleyrtp]

fixes our duplicate builds, long rebuilds, and cache thrashing by watching a specific file instead of the MANGANIS_SUPPORT env.

Since RA doesn't pass that env var, we were constantly making the entire dioxus graph recompile. Instead, you now pass a file with the env MG_CONFIG_PATH, and then manganis will use that file as its rerun-if-changed.

This is "sticky," so to unset it, you either modify that file (remove it...) or set MG_CONFIG_PATH_RESET to some value and then the build script will re-run.

I'm not a big fan of "sticky" settings but it's either this or setting the env in config.toml which occasionally does have issues and still leads to rebuilds.

[env]
mg_basepath="blah"

Alternatively, instead of the terrible CLI I grafted onto a build script, we could simply have a deterministic place to watch, but that's hard to acquire since there's no great way of getting the target dir from a proc macro.

rust-lang/cargo#8710

---

The more I think about this the more I think we should be doing one of the below:

• setting basepath in env via [env]
• exposing all manganis configuration via [env] env vars (and documenting it this way...)
• not setting basepath ever during dev and only during bundle/release
• find an alternative solution entirely
• no compile-time configuration of manganis at all (no fallbacks, always absolute routes (/a/b/c) relative to a bundle)
• use the <base> element and only relative asset uri?
• only use relative url and then build a proper directory structure?
• basepath is a runtime construct only valid via Link

also, funny, next doesn't do basepaths for images https://nextjs.org/docs/app/api-reference/next-config-js/basePath#images but does it for links