WrapperGenerator: evaluate Humanizer (or similar) so singularization scales beyond the pilot modules

[Joywambui-maina]

Problem

The singularizer in tools/WrapperGenerator/Singularizer.cs is a hand-written rule set
scoped to Graph path-segment vocabulary — irregulars (Children → Child), invariant
words (Windows), version-tagged segments (alerts_v2 → AlertV2). Every rule exists
because a shipping cmdlet name needed it, and a missing rule is only discovered when the
parity gate fails on a new module. That works at 15 pilot modules, but the rule count
scales with vocabulary as generation expands toward all ~40 modules.

Possible path forward

• Evaluate Humanizer (or similar) as the base
  inflector, keeping the existing checked-in override layer on top for Graph-specific
  behavior.
• Hard constraint: published cmdlet names must stay byte-identical. Some shipped names
  are deliberately non-standard English (onPremisesSynchronization →
  OnPremiseSynchronization) because AutoRest singularized every camelCase word — a
  correct English inflector will disagree with shipped names in places, so the override
  table must always win.
• Acceptance test: swap the library in behind the current interface and run
  tools/Compare-WrapperCmdletNames.ps1 across all pilot modules; adopt only on a
  zero-diff result.
• Also measure whether the library reduces the override table or grows it — if
  overrides grow, the dependency isn't paying for itself.