fix(cli): wheels destroy accepts <type> <name> order (#2313 F16)

[bpamiri]

Summary

• Smart-parse wheels destroy positionals so wheels destroy <type> <name> works (the form the v3 docs and wheels generate both use).
• Legacy wheels destroy <name> [type] order still recognized — no breaking change.
• Updated inline help to lead with the preferred <type> <name> form.

Closes the F16 line in #2313:

wheels destroy <type> <name> rejects the documented argument order. The CLI is treating the second positional as the type, so wheels destroy controller Posts errors with "Unknown type: posts. Valid types: resource, model, controller, view". (Tutorial currently falls back to manual rm so this is cosmetic.)

The CLI used to greet users with Unknown type: posts when they followed the docs. Now both orderings dispatch correctly:

==> Test 1: 'wheels destroy controller Posts' (was broken)
The following will be deleted:
  app/controllers/Posts.cfc
  ...
==> Test 2: legacy 'wheels destroy Post model'
The following will be deleted:
  app/models/Post.cfc
  ...

How the parse works

If the first positional matches a known type (resource/model/controller/view), treat it as the type; otherwise fall through to the legacy <name> [type] ordering. The only ambiguity case is "first arg matches a known type" — in which case there's a single sensible interpretation. Existing tests using <name> [type] keep passing because Destroyable doesn't match any known type.

Bonus catch

The first cut used inline # explanation comments in help-text examples. CFML treats # as an expression delimiter inside double-quoted strings, which crashes the whole CFC at parse time (the dreaded Invalid Syntax Closing [#] not found from the project's CLAUDE.md gotcha list). Live-loading the patched module into ~/.wheels/modules/wheels/Module.cfc surfaced this immediately — the unit-test path can't, since a broken file fails to load at all. Switched to parenthetical notes.

Test plan

• wheels destroy (no args) prints help with new <type> <name> synopsis.
• wheels destroy controller Posts --force removes app/controllers/Posts.cfc (was rejected before).
• wheels destroy Post model --force (legacy form) still removes app/models/Post.cfc.
• wheels destroy User --force (single arg) still defaults to resource.
• Three new specs in DestroyCommandSpec.cfc covering the new order.
• Full CLI test suite: 474 pass, 3 pre-existing fails in DoctorSpec > checkMixinCollisions (unrelated, predates this change).

Out of scope

The v3-0-0 destroy guide at web/sites/guides/.../core/destroy.md has separate stale issues (claims there is no --force flag, never mentions a [type] argument). Worth a dedicated docs PR alongside the broader guides rewrite — kept out of this scope to keep the diff minimal.

Other items in #2313 still outstanding (separate PRs):

• wheels start from a non-Wheels dir creates phantom servers (Polish bucket: minor CLI / framework / doc inconsistencies surfaced by VM-onboarding journals #2313 F18 partial — stop already shipped via 5e47c61).
• --help per subcommand prints top-level help (Polish bucket: minor CLI / framework / doc inconsistencies surfaced by VM-onboarding journals #2313 F18).
• SQLite migrator lower-cases column names (Polish bucket: minor CLI / framework / doc inconsistencies surfaced by VM-onboarding journals #2313 F19).
• t.timestamps() adds deletedAt without an explicit opt-in (docs vs implementation question).