Provenance forging for miri_pointer_name

[Vanille-N]

Shims to give miri_pointer_name a tag in a non-hacky way that is compatible with current Unique experiments as well as possible future GC developments.

Previous way of naming an unreachable tag was to refer to the nth parent of a reachable tag. This is very unreliable in the presence of -Zmiri-unique-is-unique which changes the number of intermediate tags between reachable tags (unique.as_ptr()) and interesting tags (unique.ptr).

A more reliable method is proposed here, with

• no more nth parent handling in miri_pointer_name itself,
• nth parent accessor still available but extracted to its own function and discouraged for n > 1 (only a macro for the direct parent),
• refering to a tag as the nearest common ancestor of two reachable tags.

This will provide the tools to easily solve the diagnostics aspect of the recent blocker for #2936

[RalfJung]

How is miri_tree_nth_parent less hacky than the previous approach? Previously at least this "nth parent" thing was only useful for naming pointers, now it lets you actually obtain that provenance in the code, which seems so much worse to me.

If the goal is just to get #2936 to land, why not just use a different depth for the two variants of the test? if cfg!(uniq) { 2 } else { 1 } or something like that should do it.