Reduce allocations in Md5Hasher and fix broken ThreadLocal<MD5> reuse

[xperiandri]

Fixes #20204

Changes

• Md5Hasher.computeHash now actually reuses the thread-local MD5 instance under NETSTANDARD2_0 (previously a new MD5 instance was created on every call, with a TODO comment noting the ThreadLocal "wasn't working"), and uses MD5.HashData on modern TFMs.
• Added Md5Hasher.hashStringInto, which hashes a string directly into a caller-supplied 16-byte Span<byte>, renting the UTF8 encoding buffer from ArrayPool<byte>.Shared instead of allocating a new byte array per call.
• Added Md5Hasher.hashStringToString, producing the hex-string hash of a string without a separate hashString |> toString allocation.
• hashString is now implemented in terms of hashStringInto.
• Updated call sites (prim-lexing.fs, FSharpProjectSnapshot.fs) to use the new allocation-reduced helpers.
• Added release notes entry.

[github-actions]

❗ Release notes required

You can open this PR in browser to add release notes: open in github.dev

---

✅ Found changes and release notes in following paths:

Change path	Release notes path	Description
`src/Compiler`	docs/release-notes/.FSharp.Compiler.Service/11.0.100.md

[T-Gro]

Nice cleanup — the ThreadLocal<MD5> is finally doing its job, and the pooled encode buffer plus the no-copy AsImmutableArray wrap remove real per-checksum allocations on a hot path.

Verified the parts that could have bitten:

• hashStringToString keeps BitConverter.ToString's dashed format, so FSharpProjectSnapshot version strings are unchanged.
• The AsImmutableArray wrap is safe here: both GetChecksum sites hand it a freshly-allocated 16-byte local that's never aliased afterwards.
• ImmutableCollectionsMarshal.AsImmutableArray is unguarded but ships in the netstandard2.0 target of System.Collections.Immutable since 8.0.0, and the repo floor is 10.0.9, so both TFMs compile.
• Md5Hasher.hashString is fully removed with no remaining callers, and Md5Hasher is internal, so no surface-area impact.

Two optional nits, neither blocking:

• The #else path ignores MD5.TryHashData's bool result. It can't fail with an exact 16-byte destination, but an assert/invalidOp on false would document that invariant.
• The XML-doc blocks are a touch long for a facilities helper; a one-liner each would read fine.