…-realsig+

A closure synthesized inside a member declared in an intrinsic augmentation
(`type C with member ...`) was emitted as a sibling of `C` in the enclosing
module class instead of nested inside `C`. Under --realsig+ a source-private
member of `C` is IL `private` (type-scoped), so the sibling closure could not
reach it and the CLR raised MethodAccessException at first invocation. Members
declared in the type's own body were always nested correctly.

GenMethodForBinding now normalizes eenv.cloc to the member's declaring type for
all non-extension members under --realsig+, so the closure nests inside the type
(idempotent for members that already had it). The program semantics are unchanged
and the private member stays IL `private`.

Fixes #19933

Co-authored-by: Copilot <[email protected]>

Co-authored-by: Copilot <[email protected]>

… shapes

Adversarial review (5 models) found the original tests under-powered: runtime-only
(no IL-nesting assertion, so they passed even on the buggy compiler under realsig-),
the instance-member case used an inlinable private that hid the bug, and most member
kinds were uncovered.

- Add Regression_RealsigAugmentationClosure_StructuralAssertions.fs: under --realsig+
  assert the closure nests under the declaring type (verifyILPresent "Type/closure@")
  and is not a module sibling (verifyILNotPresent). These FAIL on the pre-fix compiler
  and PASS on the fix, so they actually guard the IL shape.
- Mark every private member [<NoCompilerInlining>] so the call survives the optimizer
  (the previous instance test guarded nothing).
- Cover property getter/setter, indexer, static method, operator, seq, mutual rec,
  nested closures, generic method (typar threading), two-type closure-name collision,
  record/DU, nested-module and namespace-scoped types.

Co-authored-by: Copilot <[email protected]>

…lity review)

A 5-model test-quality review found the prior tests duplicated source across two
files, used a vacuous --realsig- matrix, and reinvented framework helpers.

- Merge runtime + structural files into one MemberData-driven Theory: each shape is
  asserted structurally (closure nests under the declaring type, not a module sibling)
  AND run. 23 shapes as data rows + 1 --realsig- smoke (defense-in-depth only; the fix
  is gated on realsig+, so realsig- IL is identical before/after).
- Add 3 shapes the review found uncovered and confirmed crash-on-shipped/fixed-on-PR:
  override member, secondary constructor, struct augmentation.
- Harden the two-type and async absent-fragments against the closure-name
  disambiguator ('h@N-1').
- Add an AugmentationClosureNesting.fs FileInlineData baseline (Realsig=Both) in
  Inlining.fs (idiomatic EmittedIL lock) snapshotting both the realsig+ nesting
  (closure nested in C, member IL private) and the realsig- lowering.

Co-authored-by: Copilot <[email protected]>