Fix deficiencies in meta.hasUniqueRepresentation

[SeanTUT]

Fix several errors in hasUniqueRepresentation, both false negatives and false positives. Additionally, improve documentation surrounding the function, and split its ever-growing unit test into multiple seperate tests.

• False positives resolved
  • Error sets when the backing int of the error set does not have a unique representation
  • Enums tagged by non-unique types
  • Function types
  • Pointers to comptime-only types
• False negatives resolved
  • Packed unions
  • Vectors on targets where the scalar types are bitpacked

While fixing this, I ran into several notable edge cases, so I would recommend any reviewers read the comments I left in hasUniqueRepresentation

One of the false positives in question was pointers to comptime-only types. Their representation is not defined, but to solve this, it was necessary to introduce a function to determine whether a type is comptime-only. Since this is a widely-applicable function, I made it public in meta.isComptimeOnly.
^ This has been rolled back, as determining whether a type is comptime-known takes up a massive of eval branch quota such that building any sizable project, including the compiler or even the build system, is impossible when checking this in hasUniqueRepresentation

Some rules of thumb for how hasUniqueRepresentation works now:

• False negatives can be returned, but never false positives
• Comptime-only types always return false
• Types with a well-defined layout never return a false negative or a
  false positive
• Types with no well-defined layout may return true if the type
  provably has a unique representation, but may also return a false
  negative

PS: While working on this, I ran into some bugs in std.hash.autoHash which ended up incorrectly hashing slices, optional slices, and other types which should not be able to be hashed without specifying a strategy. To resolve this, I ended up refactoring std/hash/auto_hash.zig and fixing some uses of the API which should have been compile errors.

PPS: Also while working on this, I ran into a stale reference bug in std.zig.resolveTargetQuery that was causing the CI and behavior tests to fail. To fix this, an additional allocator parameter had to be added to the function. Then, I had to correct every reference of the function within the codebase, hence the all-over-the-place changes.

[SeanTUT]

Ah, I stepped away from my computer to do something, but in a few minutes I will fix those checks and add tests for what failed

[SeanTUT]

Unfortunately, it's looking like determining if a type is comptime-only, and by extension determining whether a pointer hasUniqueRepresentation, take unseemly amount of eval branch quota. It looks like for now, we might just have to live with a false positive for pointers to comptime-only types. Language support here seems like it would be ideal.

[SeanTUT]

I've determined that the build failures come from an incorrect usage of autoHash in src/Builtin.zig, which hashes SemanticVersions (including their build/pre) by value. This exposes some problems in std/hash/auto_hash.zig which necessitate some work on that, and in addition, some functions to hash components of std.Target

[SeanTUT]

Looks like the above bug actually stems from an alarming stale pointer reference where std.zig.resolveTargetQuery returns a slice of a stack buffer used for the pre and build fields of a SemanticVersion. Adding to the fix!

[alexrp]

Looks like the above bug actually stems from an alarming stale pointer reference where std.zig.resolveTargetQuery returns a slice of a stack buffer used for the pre and build fields of a SemanticVersion. Adding to the fix!

Would you mind splitting this into a separate PR? Sounds like a fix we should land ASAP.

[SeanTUT]

For the time being, I will be closing this PR, as the tests have proven more difficult than expected, and I will likely be busy for the foreseeable future

[SeanTUT]

Additionally, there is deeper surgery to be done on related components such as auto_hash and resolveTargetQuery which are out of scope for a single PR anyways