`rustc_resolve` allocates many things in `ResolverArenas`. The lifetime
used for references into the arena is mostly `'a`, and sometimes `'b`.

This commit changes it to `'ra`, which is much more descriptive. The
commit also changes the order of lifetimes on a couple of structs so
that '`ra` is second last, before `'tcx`, and does other minor
renamings such as `'r` to `'a`.

some const cleanup: remove unnecessary attributes, add const-hack indications

I learned that we use `FIXME(const-hack)` on top of the "const-hack" label. That seems much better since it marks the right place in the code and moves around with the code. So I went through the PRs with that label and added appropriate FIXMEs in the code. IMO this means we can then remove the label -- Cc `@rust-lang/wg-const-eval.`

I also noticed some const stability attributes that don't do anything useful, and removed them.

r? `@fee1-dead`

…chenkov

Introduce `'ra` lifetime name.

`rustc_resolve` allocates many things in `ResolverArenas`. The lifetime used for references into the arena is mostly `'a`, and sometimes `'b`.

This commit changes it to `'rslv`, which is much more descriptive. The commit also changes the order of lifetimes on a couple of structs so that '`rslv` is second last, before `'tcx`, and does other minor renamings such as `'r` to `'a`.

r? `@petrochenkov`
cc `@oli-obk`

coverage: Simplify creation of sum counters

A small and self-contained improvement, extracted from some larger changes that I'm still working on.

Ultimately I want to avoid creating these sum counter-expressions in some cases (in favour of just adding physical counters directly to the nodes we care about), so a good incremental move towards that is splitting the “gather edge counters” step out from the ”build a sum of those counters” step.

Creating an extra intermediate vector should have negligible cost (and coverage isn't exercised by the benchmark suite anyway). The removed logging is redundant with the `#[instrument(..)]` logging we already have on the underlying method calls.

…mpiler-errors

more eagerly discard constraints on overflow

We always discard the results of overflowing goals inside of the trait solver. We previously did so when instantiating the response in `evaluate_goal`. Canonicalizing results only to later discard them is also  inefficient 🤷

It's simpler and nicer to debug to eagerly discard constraints inside of the query itself.

r? `@compiler-errors`