Preserve exact EDT suggestion candidates

[jackyclever]

Fixes #169.

Summary

• Retrieve case-insensitive exact-name EDT candidates independently of the fuzzy-search budget.
• Merge exact and fuzzy candidates by qualified { Name, Model } identity.
• Add regression coverage for fuzzy candidate truncation and same-named EDTs across models.
• Keep the existing public repository lookup behavior unchanged outside the suggestion path.

Root cause

SearchEdts applies ordering and a candidate limit before EdtSuggester scores results. A valid exact-name EDT outside that initial fuzzy result window was therefore never considered for the existing 1.0 exact-match score.

Impact

Exact EDT suggestions are no longer silently lost when the fuzzy candidate set is crowded. When the same EDT name exists in multiple models, each qualified candidate remains available for downstream disambiguation.

Validation

• dotnet restore d365fo-cli.slnx — passed; NuGet audit reachability and existing package advisory warnings were reported.
• dotnet build d365fo-cli.slnx --no-restore -c Release -nologo — passed with 0 errors.
• dotnet test d365fo-cli.slnx --no-build -c Release --logger console;verbosity=minimal — 1,180/1,180 passed outside the restricted sandbox.
• dotnet run --project src/D365FO.Cli -c Release --no-build -- knowledge audit --verify --output json — passed.
• dotnet run --project src/D365FO.Cli -c Release --no-build -- eval run --all --output json — 52/52 passed.
• dotnet run --project src/D365FO.Cli -c Release --no-build -- eval coverage --check --output json — passed.
• python scripts/emit-skills.py — not run locally because Python is unavailable; the upstream CI skill-drift job remains the verification gate.