Add sanity checking for query keys

[Aaron1011]

We frequently see issues where a query parameter ends up containing something that it's not supposed to (an inference variable, a placeholder region, etc): #68477 and #64964 are recent examples.

Currently, incremental compilation must be enabled to see these crashes, since they only occur when we try to hash the 'bad' type. This presents a number of issues:

1. The playground can't be used to reproduce them, since it (rightly) disables incremental compilation.
2. We may miss these kinds of issues when invoking rustc directly (e.g. the ui test suite), since -C incremental=1 is usually not passed.
3. The panic message isn't very helpful - in particular, it doesn't show the original type being hashed.

I think it would be useful to add a sanity_check method to Key, which would verify that the value is sane (e.g. no inference variances or placeholder regions) regardless of whether or not incremental compilation is enabled.

[bjorn3]

Would hashing the keys with a hasher that does nothing when not in incremental mode fix this? Because it does nothing, the call to the hasher and most code inside StableHash::stable_hash will be optimized away, but any panic will still occur.