Add benchmarks for reinterpret performance for both primitives and structs with padding.

[NHDaly]

This adds benchmarks for reinterpret performance, which is an important low-level primitive operation in Julia.

It includes benchmarks for both reinterpret on primitives, which we found to be unexpectedly slow in 1.10 but then was improved again in JuliaLang/julia@cf34aa2.

It also includes benchmarks for the new support for reinterpreting between arbitrary padded structs as long as their packedsizes are the same. We have found opportunities to improve performance for those structs, and I'll be opening a PR on julia for that. :)

These are the results for this new benchmark on julia +nightly today -- note that many of these are in ms, rather than μs:

julia> versioninfo()
Julia Version 1.14.0-DEV.1386
Commit 3b21c7f60d1 (2025-12-18 16:29 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: macOS (arm64-apple-darwin24.0.0)
  CPU: 12 × Apple M2 Max
  WORD_SIZE: 64
  LLVM: libLLVM-20.1.8 (ORCJIT, apple-m2)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 8 virtual cores)
Environment:
  JULIA_SSL_CA_ROOTS_PATH = 

julia> BaseBenchmarks.load!("reinterpret"); run(BaseBenchmarks.SUITE["reinterpret"])
4-element BenchmarkTools.BenchmarkGroup:
  tags: []
  "mixed_tuples" => 4-element BenchmarkTools.BenchmarkGroup:
          tags: []
          (104, 104) => Trial(16.791 μs)
          (228, 228) => Trial(31.916 μs)
          (0, 0) => Trial(4.166 μs)
          (100, 100) => Trial(17.209 μs)
  "padded_to_padded" => 6-element BenchmarkTools.BenchmarkGroup:
          tags: []
          (29, 48, 56) => Trial(3.970 ms)
          (10, 24, 24) => Trial(9.041 μs)
          (29, 56, 48) => Trial(3.756 ms)
          (117, 128, 128) => Trial(17.475 ms)
          (10, 12, 24) => Trial(131.583 μs)
          (0, 0, 0) => Trial(4.250 μs)
  "packed_types" => 5-element BenchmarkTools.BenchmarkGroup:
          tags: []
          17 => Trial(9.459 μs)
          49 => Trial(12.708 μs)
          8 => Trial(12.042 μs)
          128 => Trial(19.292 μs)
          0 => Trial(4.125 μs)
  "padded_types" => 3-element BenchmarkTools.BenchmarkGroup:
          tags: []
          (29, 56) => Trial(3.720 ms)
          (10, 24) => Trial(73.292 μs)
          (117, 128) => Trial(99.875 μs)

EDIT: This benchmark is used in the upstream Julia PR here: JuliaLang/julia#60415