incremental: refactor out non-query data from tcx

[nikomatsakis]

There are various fields in the tcx of data that is not tracked for incremental compilation, or is tracked in an odd way. Here is a partial list:

• used_mut_nodes - Refactored used_mut_nodes #43945
• rvalue_promotable_to_static - remove or encapsulate the remaining non-query data in tcx #44501
• normalized_cache - remove or encapsulate the remaining non-query data in tcx #44501
• inhabitedness_cache - remove or encapsulate the remaining non-query data in tcx #44501
• trait_map - rustc: Make the trait_map of TyCtxt private #44162
• export_map - rustc: Make the trait_map of TyCtxt private #44162
• named_region_map - Migrate a slew of metadata methods to queries #44142
• specializes_cache - rustc: Remove specialization_cache in favor of a query #44157
• freevars - Migrate a slew of metadata methods to queries #44142
• maybe_unused_trait_imports - Migrate a slew of metadata methods to queries #44142
• maybe_unused_extern_crates - Migrate a slew of metadata methods to queries #44142
• used_unsafe - rustc: Remove the used_unsafe field on TyCtxt #44195
• stability - Migrate a slew of metadata methods to queries #44142

In each case, the idea would be to (at least to start) encapsulate the fields so that we are using queries to access those maps. I'll try to write up some mentoring instructions for each one later on.

cc @michaelwoerister

[petrochenkov]

Regarding named_region_map, I wanted to move lifetime resolution to AST some time, and then store lifetime resolutions directly in HIR (like resolutions for other names/paths), including various late-bound and elided lifetimes.

[nikomatsakis]

Here are some other fields that look suspicious to me:

list omitted and moved to the root of the issue

Many of these follow the pattern of "maps produced by resolve that ought to be accessed via queries and ought not to be indexed by NodeId".