Library sync and cache invalidation

[castwide]

• Library methods that use cursors need to sync_catalog first.
• Ensure that changing a method's return type in YARD tags resets the inference cache.
• Ensure that changes to local pins in source maps reset the inference cache.
• Host attaches opened and unmerged sources to libraries

[Copilot]

Pull Request Overview

This PR addresses improvements to library syncing and cache invalidation by updating the way source maps are merged and refreshed, as well as refining how local pins and catalog synchronization are handled. Key changes include the removal of several tests for source map merge behavior in the specs, refactoring of the source map and library code to use a sync counter instead of thread management, and adjustments to the parsing and finalization logic in source processing.

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
spec/source_map_spec.rb	Removed multiple tests for merging behavior
spec/pin/method_spec.rb	Minor stylistic changes with adjustments in test descriptions
spec/language_server/host/message_worker_spec.rb	Update of message double construction and synchronization
lib/solargraph/source_map/data.rb	New data wrapper for lazy generation of pins and locals
lib/solargraph/source_map.rb	Refactored source map initialization and merging logic
lib/solargraph/source.rb	Revised finalization and synchronization of source parsing
lib/solargraph/pin/method.rb	Updated merging logic in nearly? method
lib/solargraph/library.rb	Replaced thread management with a sync counter
lib/solargraph/language_server/host/message_worker.rb	Introduced message prioritization logic in worker class

Comments suppressed due to low confidence (1)

spec/source_map_spec.rb:74

• Removal of several tests for merging behaviors reduces coverage on source map merging functionality; consider adding replacement tests to validate that the intended merge behavior is still correctly handled.

  -  it "merges comment changes" do

[castwide]

The three major bottlenecks in library updates are source parsing, source mapping, and api_map cataloging. The top-level approach for improving them is to apply changes at the moment they're needed instead of when they're received.

• Updating (synchronizing) sources creates a new source that tracks the changes without applying them. The library will parse them when a query requires the updates to be executed. This allows sources to queue changes from the language server (e.g., via textDocument/didChange) instead of triggering a parse for each change.
• Source maps will not generate pins until they're required by a library catalog.
• Source maps used to perform a #try_merge! to determine if a source map's pins could be merged into an existing map to reduce cataloging. Any benefit from this process was lost to the fact that the source map needed to be generated before it could be merged. JIT mapping in the library is ultimately more efficient.
• We can explore whether try_merge! would be more effective as part of the Library#sync_catalog process, where it still might reduce ApiMap#catalog overhead. The question is how much the reduction in cataloging compares to the addition of merge attempts.