Optimize Query.Newline + reduce query/region allocations

[dsisco11]

Summary
This PR makes Query.Newline both semantically precise (“node occurs after a newline”) and fast by relying on green-node boundary flags instead of scanning trivia. It also removes a small set of high-impact LINQ usages from hot query paths and region resolution to reduce allocations and improve throughput.

What Changed

• Correct + optimize newline matching:
  • Defines newline semantics as: current node has leading newline trivia OR previous sibling has trailing newline trivia.
  • Implements green matching using boundary flags only (no red-node allocation, no trivia scans).
• Fix flag propagation semantics (token-centric ownership):
  • Boundary trivia flags are owned by tokens, not containers; containers track “contains” flags but do not claim boundary flags.
  • Adjusts propagation for lists/syntax containers to avoid incorrect boundary flag inheritance.
• Reduce allocations in the query engine:
  • Replaces LINQ in selection modes and hottest Select(...)/combinator paths with tight loops to avoid iterator/closure allocations.
• Speed up SyntaxEditor region resolution:
  • Refactors region traversal to keep an incremental slot/path stack and only materialize NodePath when a match is found.
• Adds coverage + perf harness:
  • Adds unit tests for newline semantics (top-level + inside blocks) and verifies Query.NotNewline is the exact negation under the same context.
  • Adds SyntaxEditor mutation tests to ensure green flags stay correct across insert/replace/remove and undo/redo.
  • Adds a targeted BenchmarkDotNet suite for Query.Newline.

Key Files

• Newline + query engine changes: NodeQueryTypes.cs, QueryCombinators.cs, QueryRegion.cs
• Flag semantics + propagation: GreenNodeFlags.cs, GreenLeaf.cs, GreenList.cs, GreenSyntaxNode.cs
• Tests: NewlineQueryTests.cs, SyntaxEditorTests.cs, GreenNodeFlagsTests.cs
• Benchmarks: NewlineQueryBenchmarks.cs, Program.cs
• Tracking doc: newline-query-optimization.todo

Benchmarks (NewlineQueryBenchmarks)
Using fixed settings (--warmupCount 3 --iterationCount 30 --outliers RemoveAll), current branch vs master:

• Select(Query.Newline) - Count
  • Lines=1000: ~9.7% faster, alloc essentially unchanged
  • Lines=10000: ~5.6% faster, alloc essentially unchanged
• Select(Query.Newline.First()) - First match
  • Lines=1000: ~13.6% faster, ~6.7% less allocated
  • Lines=10000: ~15.1% faster, ~6.7% less allocated