fix: resolve JVM imports in Gradle/Maven module source roots

[ariancovac]

Problem

Multi-module Gradle/Maven projects keep each module's sources under a module directory (android/src/main/java, feature/x/src/main/kotlin). The Java and Kotlin import resolvers only tried scan-root-relative paths (src/main/java, src, app/src/main/java, .), so in any multi-module project every cross-file import failed to resolve:

• the import graph came back with zero edges,
• importer_count == 0 for essentially the whole codebase,
• graph-based detectors (orphaned, coupling, cycles, single_use, test coverage) flagged the entire project — e.g. the orphaned detector reported 118 "orphaned" files including MainActivity, MediaKitApplication, and settings.gradle.kts.

Verified against a real 3-module Android project (modules android/, download-private/, detekt-rules/): before the fix, 118/120 source files were flagged orphaned; after the fix, 0.

Fix

• Add discover_jvm_source_roots(scan_path) to resolver_cache.py: an os.walk (pruning build/, node_modules/, dot-dirs, depth-capped) that finds <module>/src/main/java|kotlin roots, memoized per scan root and cleared by reset_import_cache().
• resolve_java_import / resolve_kotlin_import now try the discovered module roots first, then the historical fallbacks (behavior for single-module/flat layouts is unchanged).
• New test test_jvm_resolvers_find_imports_in_gradle_modules covers Kotlin + Java module resolution, ensures build-output copies never win over real sources, and exercises the cache reset.

Full suite: 601 passed in tests/lang (the 2 failures in test_treesitter_imports_direct.py — test_backend_import_resolvers_cover_language_specific_paths, test_script_import_resolvers_cover_cached_paths_and_extensions — also fail on clean main on Windows; pre-existing, unrelated).

[awdemos]

Reviewed against current main — the fix is real and well-built (exclusion-aware discover_jvm_source_roots walk, depth-capped, lru-cached, cleared in reset_import_cache), but one thing needs correcting before this is mergeable:

The existing test failure is introduced by this PR, not pre-existing. test_backend_import_resolvers_cover_language_specific_paths fails on this branch (1 failed, 9 passed) and passes on clean main on Linux. The body claims it "also fails on clean main on Windows" — I could not reproduce that. Mechanism: _jvm_candidate_roots drops the historical src/main/kotlin fallback that resolve_kotlin_import had, so when an earlier resolve_java_import call memoizes discovered roots before the test creates the kotlin dir, the lookup returns None.

Restoring os.path.join(scan_path, "src", "main", "kotlin") to the fallback list makes all 10 tests pass.

Follow-up (non-blocking): reset_import_cache() is only called from tests, so in a long-lived process the discovery cache can go stale across scans — worth wiring into the scan lifecycle.

Also note src/jvmMain/kotlin-style KMP main source roots are still not discovered (this covers Gradle/Maven src/main/java|kotlin); pairs naturally with #677 which handles the KMP test side.