Rotate [<return: X>] attributes during binding normalization

[nojaf]

Description

Follow-up to #19738. That PR moved [<return: X>] attributes written in front of a binding out of SynBinding.attributes and into SynValInfo.retInfo, and it did so in mkSynBinding, in the parser. This PR keeps both fixes from that change but performs the rotation one layer later, so the untyped tree again describes the source.

Given:

[<return: Struct>]
let (|Foo|_|) x = ValueNone

the binding reports no attributes for source that visibly has one. Routing the attribute to the return-value metadata slot is right for the type checker, IL emit and the Symbols API, but SynBinding is not one of those consumers; its contract is to say where the user wrote it.

The rotation is not new. It previously lived in TcNormalizedBinding and patched a local valSynData, so the parse tree stayed faithful to the source. Doing the same work in mkSynBinding hands every untyped-tree consumer a tree that no longer matches what was written: formatters, analyzers, source generators, refactoring tooling.

It was also lossy in two ways that cannot be recovered downstream:

• The attribute list range narrowed from the [< >] span to the attribute alone, so the brackets the user typed were no longer represented anywhere.
• RotateReturnAttributes collected every return attribute into one synthesized list, so [<return: A; return: B>] and [<return: A>][<return: B>] produced identical trees. Neither can be printed back to its original form.

Fantomas hit this while moving onto a newer FCS (fsprojects/fantomas#3400): partial active patterns marked [<return: Struct>] were parsed and then never printed, silently deleting them from every formatted file, including 34 in Fantomas' own source.

The change

Move the rotation to BindingNormalization.NormalizeBinding, the single funnel from SynBinding to NormalizedBinding and already a lowering step. NormalizedBinding holds a flat SynAttribute list, so RotateReturnAttributes now takes and returns that instead of SynAttributes, and the list-splicing that destroyed grouping is gone with it.

Everything downstream (TcNormalizedBinding, AnalyzeAndMakeAndPublishRecursiveValue, the object-expression paths) consumes NormalizeBinding's output, so TcNormalizedBinding keeps reading retInfo as the single source of truth and the ValReprInfo lookup in NameResolution for [<return: Struct>] active patterns is untouched.

The return type annotation form (let f x : [<return: A>] int = x) never went through this rotation; those attributes are placed in retInfo by InferSynReturnData and are unaffected.

SynBinding now again carries the attribute, with the full [< >] range, and retInfo is empty at parse time:

SynBinding
  (None, Normal, false, false,
   [{ Attributes = [{ TypeName = SynLongIdent ([Struct], [], [None])
                      Target = Some return
                      Range = (3,2--3,16) }]
      Range = (3,0--3,18) }],
   ...
   SynValInfo ([[SynArgInfo ([], false, Some x)]], SynArgInfo ([], false, None))

The two grouping forms are distinguishable again: [<return: A>][<return: A>] yields two SynAttributeLists, [<return: A; return: A>] yields one.

Fixes #: not applicable, no filed issue; this is a follow-up to PR #19738.

Checklist

• Test cases added
• Performance benchmarks added in case of performance changes
• Release notes entry updated:

  docs/release-notes/.FSharp.Compiler.Service/11.0.100.md, under ### Changed.

[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

[nojaf]

@T-Gro ready for review.
CI failure seems unrelated.

//cc @edgarfgp @auduchinok

[T-Gro]

Could you also pls trigger the "post mortem" skill so that we generate instructions that will prevent similar breaks in the future?

[nojaf]

@T-Gro thanks, done!

[github-actions]

🔍 Tooling Safety Check — Affects-Agent-Config, Affects-Compiler-Output
Affects-Agent-Config: modifies .github/instructions/SyntaxTree.instructions.md
Affects-Compiler-Output: changes SyntaxTreeOps parse tree structure

Generated by PR Tooling Safety Check · opus46 4.4M · ◷

[nojaf]

@T-Gro I think this MacOS Batch1 failure is unrelated:

Let me know if I need to do something here.