Fluent, with all the icu4x it brings in, takes quite some time to
compile. `fluent_messages!` is only needed in further downstream rustc
crates, but is blocking more upstream crates like `rustc_index`. By
splitting it out, we allow `rustc_macros` to be compiled earlier, which
speeds up `x check compiler` by about 5 seconds (and even more after the
needless dependency on `serde_json` is removed from
`rustc_data_structures`).
This allows analyzing the output programatically; for example, finding
the item with the highest `total_estimate`.
I also took the liberty of adding `untracked` tests to `rustc_session` and documentation to the unstable book for `dump-mono-items`.