Thirteen rule sets describe the rules they run

[Rafael-SOWNet]

RewriteRuleSet.Rules is what the registry reports a set is made of, and it came from
RuleRegistryGenerator reading the switch that defined the set. Twenty-seven of the thirty sets
stopped running that switch some releases ago — they run MatchedRuleSet.ApplyHere — and went on
describing it. That is #825's open half, and what #746 tier 2 names as gating the rest.

Thirteen of them now describe what they run. #1107 made that possible by giving AsAddressable() the
rule's exact growth, its tier, and somewhere for its identity to live.

Which thirteen, and why those

CollapseMultipleFractions, the three CommonDenominator sets, DivisionPreparing,
ExpandFactorialDivisions, ExpandMultipleAngle, ExpandTrigonometric, Expansion,
FactorizeFactorialMultiplications, NormalTrigonometricForm, PhiFunction,
PolynomialLongDivision.

Not one of them had a single described arm. The generator only has a description where somebody
wrote a comment above the arm, and for these nobody had — so repointing them adds metadata rather
than trading it. The sets left alone would trade: Common would lose 33 descriptions today and
Power 22. Porting those is the next change, not this one.

The identities come from the comments the data rules already carry, checked against each rule's own
pattern and replacement rather than copied — a * (1 / b) = a / b,
(a / b) / (c / d) = (a * d) / (b * c), tan(a) = sin(a) / cos(a). Four of the eight
CollapseMultipleFractions rules had no comment at all and their identities were read off the
patterns. Thirty-two rules gain one, and across the registry the number of described rules goes from
95 to 135.

What changes for a caller

Measured on a build of each version rather than read off the diff, and recorded in
BREAKING-CHANGES.md.

	Was	Is
Rules[i].Name	the arm's rendered pattern, Mulf(var any1, Divf(Integer(1), var any2))	the rule's name, reciprocal-factor-becomes-a-quotient
Rules.Count, two factorial sets	8	3
RewriteRules.All.Sum(set => set.Rules.Count)	407	397
Rules[i].Description	null, for all 38 of these arms	the identity, for all 38
Rules[i].Soundness	null — an arm declares no tier	the rule's own tier
Rules[i].PatternSource	the C# the arm was written in	the pattern the matcher holds

The two counts that fall are not rewrites lost. ExpandFactorialDivisions and
FactorizeFactorialMultiplications are eight arms each that the data form writes as three: the other
five are the same rewrite spelled once for each side a factorial can sit on, which is one commutative
pattern. Every other repointed set is one arm to one rule.

One growth becomes less certain, and that is the right answer

ExpandFactorialDivisions[0] reported Collects. That was the string-length proxy guessing. The
exact answer for a code-built rule nobody declared is Unknown — and Unknown is right here: a
quotient of factorials collects when the offsets are one apart and expands when they are five, so
no single value is true of the rule.

The two ExpandMultipleAngle rules went the other way. The proxy had guessed Expands and happened
to be right, so the exact Unknown would have lost real information; they are now declared
Expands, since the Chebyshev expansion of sin(n * a) is a sum of n terms where the pattern is one
node, for every n the rule fires on. That is what RewriteRuleGrowth.Unknown exists to prompt.

The test suite is what found this. GrowthSaysWhichWayTheSetMoves failed on ExpandMultipleAngle
the moment the proxy's guess was gone, which is how the declaration got written rather than the
assertion got relaxed.

What is untouched

RuleConfluenceTest, DerivationPathTest and RewriteRecordingTest all pass unchanged — they are
about Common and Power, which this does not repoint. Three tests changed: the two registry counts,
and my own RuleMetadataTest list of which sets carry a per-rule tier, which is now named rather than
counted so that repointing a set is a visible change.

Still explicitly not here: deleting the ~27 dead switch methods. They are the oracle
MatchedRulesAgreeWithTheSwitchTest proves the data form against, so deleting them deletes the
evidence the exchange was safe. That is a maintainer decision for #825.

State

Full suite: 8974 passed, 14 skipped, 0 failed. No public API change.

Part of #746 tier 2 and #825.