Match LIFT list values ignoring surrounding whitespace - #1200

Summary

FieldWorks semantic domain templates before 2017 have trailing spaces in 16 English names ("Necessary ", "Animism ", ...), and projects created then still have them. The Combine writes the trimmed 9.4.2.3 Necessary to LIFT (sillsdev/TheCombine#4410). That misses FLEx's key 9.4.2.3 Necessary , so import adds a duplicate domain.

 TryGetPossibilityMatchingTrait(value)
-  try value, lower(value)
+  TryGetPossibilityMatchingUntrimmed(value)     // value, lower(value)
+  else TryGetPossibilityMatchingTrimmed(value)  // trimmed, lower(trimmed);
+                                                // else scan for key.Trim() == trimmed

-FindComplexFormType(old entry-type) / FindVariantType(old entry-type)
+FindOldEntryType(old entry-type)
+  untrimmed complex form type, else untrimmed variant type,
+  else trimmed complex form type, else trimmed variant type

The maps are unchanged, so an untrimmed match always wins. The scan runs only after every direct lookup misses; hits aren't cached, so a value that matches only after trimming rescans on each occurrence.

Evidence

New tests in LiftMergerTests cover padded names in semantic domains, parts of speech and both entry-type lists, the precedence rules above, whitespace variants within one file, minor-entry-condition subtypes, and the no-match fallbacks.

• Before: ImportSemanticDomainTrait_MatchesExistingDomainWithPaddedName on main: Should not create a new semantic domain. Expected: property Count equal to 1 But was: 2
• After: LiftMergerTests and LiftMergerRelationTests: 79 passed, 0 failed, 2 skipped (TestLiftImport9C/9D, which skip themselves).

Merge Danger

Door: two-way. Import-time matching only; no model or data migration.

Blast Radius: LIFT import. The trimmed fallback applies to every list matched through TryGetPossibilityMatchingTrait (semantic domains, parts of speech, entry types, anthropology codes, dialects, usage types, locations and the rest), not just semantic domains.

Out of scope, still matched without trimming: reversal parts of speech, lexical relation types, exception features, morph types, list-valued custom fields, minor-entry-condition subtypes, and the old entry-type special cases (Derivation, main entry, ...).

Considered and rejected

• Storing trimmed aliases in the possibility maps. An alias could claim a key before a later item whose name matched exactly.
• Matching lift-ranges semantic domains by label. ProcessSemanticDomain matches range elements only by guid. That's separate, pre-existing behavior.

🤖 Generated with Claude Code

This change is