powershell-lsp: close the I0.1 rationale residual (ManifestConsistency) + v1.24.1 release-prep (000124)

[manderse21]

Closes the I0.1 rationale-coverage residual and cuts v1.24.1. Dispatch 000124, leg 1 (leg 2 is a
survey and contributes no diff). PATCH-bearing, 8 files, +159/-26.

What changed

• The entry. scripts/regen-rule-rationales.ps1 gains a fifth hand-authored owned rationale,
  keyed ManifestConsistency -- the ruleId the finder actually emits (lsp-common.ps1:2196,2208,
  pses-daemon.ps1:816), confirmed on disk before authoring, not the function name
  Test-ManifestConsistency. 175 chars against the 180 cap, pure printable ASCII.
• The table. rulesets/rule-rationales.psd1 regenerated by the script, never hand-edited:
  59 entries = 54 auto-derived PSSA (unchanged, byte-for-byte) + 5 owned. The artifact diff is
  3 insertions / 1 deletion.
• The text is faithful to the predicate. One code carries two finding kinds, so the rationale
  names both: a manifest listing a function that is never defined (orphan-export), and a function the
  module defines and exports that FunctionsToExport omits (under-declared-export). It does not claim
  the module is broken, because every indeterminate shape (wildcard '*', empty/null
  FunctionsToExport, dot-source, dynamic export, unparseable module) degrades before any finding
  is emitted.
• Guards moved to the 5-owned world. Enumerated from disk rather than assumed -- there were
  three, not one (the count assertions read the table from disk and self-adjust).
• Roadmap. The Horizon 0 I0.1 residual clause is retired; the entry reads closed.

A correctness fact this surfaced

ManifestConsistency is emitted from a second site for the indeterminate case
(pses-daemon.ps1:816, _indeterminate = $true, "cannot be determined"). lsp-client.ps1 routes
those to the prose-note channel and only determinate findings into $diags, so the new entry can
never render "the lists disagree" onto a note saying the opposite. That separation was previously
unguarded and is now pinned by an integration test that goes RED if it is removed.

Decisions (the inbox's open questions)

• OQ1 -- one PR. Leg 2 produces no diff, so this carries leg 1's build + docs + release commits
  and one merge gate (the 000121 shape).
• OQ2 -- PATCH, v1.24.1. The versioning policy in CHANGELOG.md reserves MINOR for a new
  capability. The rationale feature shipped at v1.24.0; this completes the coverage of its generated
  data table. Decisive on-disk precedent: [1.21.1] shipped as PATCH while removing three rules
  from the base surface -- a strictly larger observable-diagnostics change than adding one prose
  line. No deviation.
• OQ3 -- no compatibility note. The 000121-era verified outbox custom_checks that assert the
  4-owned world are terminal and not re-run. The live guards that pinned 4 were in scope and are
  updated.

Deviation (approved in-session by Mike)

The Section 3 feedback table, row 9, still read one-entry residual -> I0.1, which is outside the
dispatch's named scope (scope_in names Horizon 0; scope_out forbids other roadmap edits). Left
alone it would make the roadmap self-contradict -- Horizon 0 closed, table row open. Mike granted the
one-cell edit explicitly. The roadmap diff is exactly two hunks: the named clause and that cell.

Verification

Every check proven non-vacuous, each by RED-then-GREEN, with the artifact restored bit-for-bit
after each mutation:

• regen -Check exits 0 on the shipped table; exits 1 on a perturbed rationale text and on a
  dropped owned-list entry, each with a precise diff.
• The three updated unit guards go RED against the pre-change (4-owned) table restored from HEAD;
  the degrade guard, which is invariant across this change, was instead proven RED by injecting an
  entry for its exemplar.
• The new integration assertion goes RED on the pre-change table -- the finding renders with no
  why: line, which is precisely the behaviour this closes.
• The degrade path keeps a real exemplar rather than a synthetic one: PSUseSingularNouns is a
  genuine PSScriptAnalyzer code, deliberately entry-less because 000092 excluded it from base.psd1
  as measured-noisy, yet still reachable when a user's own PSScriptAnalyzerSettings.psd1 enables it.
  The test asserts it is absent from the table so the exemplar cannot silently stop being one, and
  says so in a comment for future curation / pin-bump work. The synthetic case is kept as a second
  assertion.

Full local suite green on both hosts: pwsh 7 -- 1194 passed, 0 failed; Windows PowerShell 5.1 --
1192 passed, 0 failed, 2 skipped. Frozen properties hold: a clean file still emits nothing, and
per-rule dedup is unchanged.

Tag and release pipeline are untouched -- bump-version.ps1 only prints the tag command. HELD for
Mike's merge gate.

https://claude.ai/code/session_01LzLuvSXsCK7ahhK1WUCFGT