Remove HermitCrab's analysis-cascade memo

[johnml1135]

Quick summary

Removes HermitCrab's opt-in analysis-cascade memo; the sequential rule cascade
and MergeEquivalentAnalyses still work exactly as before. Measured as a
package (#490), it cost 28-32% and ~29 GB on Sena for no net speedup on FLEx's
parallel path. Pruning (#474 landed; #491/#494/#519 open) is untouched and is
where the real parse-time wins come from.

Where to look

• AnalysisMergeKey (AnalysisStratumRule.cs L226-309) still distinguishes non-head count and
  per-rule unapplication counts -- pinned by AnalysisMergeKeyTests.cs and MergeEquivalentAnalysesTests.cs.
• Memo plumbing is gone (AnalysisScope.cs, AnalysisStateKey.cs, MemoizedCombinationRuleCascade.cs
  deleted) -- maxDegreeOfParallelism and the sequential cascade are unchanged public API.
• Review guidance dropped the memo-key re-audit language (docs/review/hermitcrab.md and others).

Validation

• dotnet build (MSBUILDDISABLENODEREUSE=1) -- succeeded, 0 warnings, 0 errors.
• dotnet test tests/SIL.Machine.Morphology.HermitCrab.Tests --no-build -- 89 passed, 0 failed.
• dotnet csharpier check . -- 713 files checked, clean.
• pwsh -NoProfile -File scripts/comment-hygiene.ps1 -BaseRef origin/master -- clean.
• git diff origin/master..HEAD --stat -- 16 files, +196/-1520; production diff unchanged, only
  AnalysisMergeKeyTests.cs added.
• ./local_check.sh was not run as such; the commands above cover its steps individually.

---

Reading this a year from now

The memo only ran when maxDegreeOfParallelism == 1 and tracing was off. FLEx uses
the default parallelism, so production never executed it. #490's ledger measured
the optimization round as a package: no net speedup, and cross-word memo sharing
was 28-32% slower on Sena and cost ~29 GB of memory. The gains keep coming from
search-space pruning that doesn't need a replay cache: #474 (landed), and #491,
#494, and #519 (open at the time of writing). Those shrink the work instead of
caching it, and they apply on the parallel path FLEx actually uses. The memo was
~1,400 lines of subtle code (ReplayOnto grafting, arrival-count bookkeeping)
whose correctness depended on an unenforced key invariant (#490, "narrowing that
key breaks correctness with no test failure").

What stays, and why

• The public Morpher(..., maxDegreeOfParallelism) parameter and the sequential
  cascade. They are API and still useful without the memo.
• Fix bug in MergeEquivalentAnalyses #493's MergeEquivalentAnalyses fix. It keyed on AnalysisStateKey; that key
  now lives as an internal AnalysisMergeKey in AnalysisStratumRule, covering the
  same dimensions (shape, stratum, syntactic and realizational feature structures,
  non-head count, per-rule unapplication counts).

Prior measurements (not reproduced by this review round)

These are earlier author measurements from the memo-removal work, not commands run
as part of this pass's validation, and no reproducible artifact for them is checked
into the tree:

• Spot check on 5 real grammars (Sena, Amharic, Indonesian, Mbugwe, Aweti), 5 words
  each, all with at least one parse: analysis multisets (gloss, allomorph identity,
  root index, feature structures) were reported identical between master memo-off
  and this branch in both default and sequential modes, 25/25 match.
• On the 14 words where master memo-on also completed, it was reported no faster
  than this branch's sequential mode (e.g. 3,286 ms memo-on vs 3,122 ms without it;
  412 vs 350 ms; the rest within noise).
• Full-solution run reported as 1,003 passed / 3 skipped / 0 failed.

🤖 Generated with Claude Code

---

This change is 

[codecov-commenter]

Codecov Report

❌ Patch coverage is 75.00000% with 21 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.90%. Comparing base (7109531) to head (a84dc6f).

Files with missing lines	Patch %	Lines
...chine.Morphology.HermitCrab/AnalysisStratumRule.cs	74.69%	11 Missing and 10 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #522      +/-   ##
==========================================
- Coverage   74.07%   73.90%   -0.18%     
==========================================
  Files         456      453       -3     
  Lines       38164    38000     -164     
  Branches     5228     5206      -22     
==========================================
- Hits        28270    28083     -187     
- Misses       8734     8751      +17     
- Partials     1160     1166       +6     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.