Scala baseline parity, then SemanticDB semantic facts

[vitali87]

Why

Scala is the emptiest slot in the language matrix: no LanguageHandler registration, no import parsing, no type inference, no parsers/scala/ package — only FQN node-type tuples in constants/fqn_specs.py:363-508 and a single infix-operator special case at call_processor.py:2616. The support matrix marks it In Development.

Scala is also the language where heuristics have the lowest ceiling: implicits, extension methods, and for-comprehension desugaring are compiler transformations — no syntactic analysis will ever resolve them.

Plan — two stages

Stage 1: baseline tree-sitter parity

Bring Scala up to what every other language already has:

• LanguageHandler registration + parsers/scala/ package
• Import parsing: import a.b.c, wildcard _/*, selector groups and renames (import a.{B => C}), Scala 3 given imports
• Definitions: object (and its companion pairing), case classes, traits, given/extension (Scala 3), nested defs
• INHERITS/IMPLEMENTS from extends ... with ... clauses (trait linearization at least recorded, even if unresolved)

This unblocks the Scala FLOWS_TO descriptor issue and makes the existing scala_retrieval eval meaningful.

Stage 2: SemanticDB semantic facts

SemanticDB is purpose-built for exactly this use case — it's the format Metals runs on, and evals/oracles/scala_oracle/ already consumes scalameta:

• Scala 3 emits it with a compiler flag (-Xsemanticdb); Scala 2 via the semanticdb-scalac compiler plugin.
• Each .semanticdb file contains resolved symbol occurrences for every identifier: implicit resolutions, extension-method binding, desugared for-comprehensions, overload selection.
• Adapter: parse the protobuf payloads, map occurrences to LanguageFrontend fact families (resolved_call_sites, external_sites, base_kinds by symbol, expression_types from signatures).

Friction: SemanticDB requires a compile via sbt/bloop/mill, so stage 2 is strictly opt-in — point CGR at the build's META-INF/semanticdb output directory, or run sbt compile behind a flag. A missing build degrades to stage 1 behaviour.

Tests

Fixture Scala repo per stage: stage 1 — imports with renames, companion objects, trait mixins; stage 2 — an implicit conversion, an extension method call, a for-comprehension over a custom monad, overloads. Cross-check calls against run_scala_call_oracle.

Stage 2 depends on the LanguageFrontend protocol issue.