[Draft] Move query definitions into rustc_query_system

[Julian-Wollersberger]

... and out of rustc_middle, to improve bootstrap times. (cc #65031)

The goal here is to allow future simplifications of the query system, and to make rustc_middle smaller, so that x.py check and x.py build is more parallel.

The important part are the first two commits:
The query definitions in the rustc_queries! proc macro generates several macro_rules macros that are invoked in different parts of rustc_middle. The change here is that one of them, query_description_stream!, now gets all the needed use statements as an input, rather than relying that they are defined at the top of the file. This makes it possible to move the query definitions into rustc_query_system although types like HirId or DefId aren't available there.

The rest of this PR demonstrates one simplification that this schema enables:
DepKind doesn't depend on the query parameter types that are not accessible in rustc_query_system , so one of the macros generated by rustc_queries! can be invoked in rustc_query_system directly to generate the DepKind enum. This enables removing (most of) the DepKind trait and thus removing the generic argument in a bunch of places. (Basically every <D> in #77871.) This doesn't fully work yet, because eg. DepKind::debug needs the global ImplicitTyCxt.

I guess this is an alternative approach to #70951, or maybe a first step to that. (My next todo is reading that PR.) CC rust-lang/compiler-team#277
Can I get your opinions on this approach, @cjgillot and maybe @eddyb and @oli-obk ? Is this a direction worth pursuing?

This PR is more a proof-of-concept and needs cleanup:

• Update docs to the new location for query definitions
• do git move on the query definitions file to not lose history
• rebase on Make fewer types generic over QueryContext #77871
• make the refactoring of DepKind in a separate PR
• measure how long rustc_queries! takes, so we know the actual improvement here.

[rust-highfive]

r? @oli-obk

(rust_highfive has picked a reviewer for you, use r? to override)

[Julian-Wollersberger]

@rustbot modify labels: +A-query-system +I-compiletime +T-compiler

[jyn514]

do git move on the query definitions file to not lose history

This will have no effect, git move is the same as deleting the old files and running git add on the new ones. GitHub is just bad at following renames.

[cjgillot]

You can surround thins expansion by an ad-hoc module to avoid polluting the namespace.

[cjgillot]

FWIW, when I created rustc_query_system, I wanted to have the DepGraph parametered by the DepContext, instead of the DepKind. This required to have DepGraph<TyCtxt<'_>> with a lifetime parameter, which caused lifetime variance problems. The read_deps and with_deps functions in DepKind trait are the middle ground I found at the time.

[Julian-Wollersberger]

I currently don't have much time to work on this, so I'll close it for now.
And university starts again next week, so I won't have time in the foreseeable future.

[Julian-Wollersberger]

measure how long rustc_queries! takes, so we know the actual improvement here.

FWIW, the proc macro takes 350ms to execute.
(Measured with eprintln!("Query proc macro took {:?}.", end_time.duration_since(start_time));)
The total x.py check time is 2sec less, but that's the variance I normally get.