perf: unfold `binderNameHint` eagerly in the kernel (#14809)

This PR unfolds `binderNameHint` eagerly in the kernel. The definition
becomes an `abbrev` and the `.abbrev` reducibility hint makes the
kernel's lazy delta unfold the gadget earerly. Explicit `attribute`
lines restore the `[implicit_reducible]` transparency and `simp ↓`
attribute, so elaboration and tactics keep treating the gadget as
before.

What problem does `.abbrev` fix? Consider
```lean
myHint 0 (fun n => n) (W 100000)  =?=  W 100000
```
Here, `isDefEq` and the kernel both see a constant applied to arguments
and must make a choice which constant to unfold first. For
`ReducibilityHint.regular`, the choice is to unfold `W` because it has
greater *height* than height 1 for `myHint`. And then this problem
recurs on the order of 100000 times. By contrast, for `.abbrev`,
`myHint` is chosen and the defeq succeeds in constant time.

This caused quadratic kernel checking time (and space) in #14590 and
likely would do so in other `binderNameHint`-heavy benchmarks.
2 files changed