perf(editor): eliminate quadratic line counting and substring allocations in SqlStatementSplitter (#1350)

[ZhuchkaTriplesix]

Closes #1350

Problem

1. When parsing large SQL files, line counting in SqlStatementSplitter previously incurred quadratic scans O(N*M) and sql.substring(i) string allocations with dynamic RegExp creation on every dollar sign.
2. Synchronous character-by-character scanning with sql[i] allocated individual character string objects across millions of characters.
3. Splitting multi-megabyte scripts ran synchronously on the UI thread, risking dropped frames on query execution.

Solution

• Kept the linear newline scan and static dollar tag regex.
• Switched the scanner from sql[i] substring indexing to sql.codeUnitAt(i), eliminating all intermediate character allocations.
• Added SqlStatementSplitter.spansAsync which offloads statement splitting to a background isolate via compute for scripts exceeding 64 KB (defaultIsolateThreshold).
• Switched GenericSqlWorkspace._runSpans to async, leveraging spansAsync to protect UI responsiveness during script execution.
• Added unit tests covering spansAsync (both sync below threshold and background isolate above threshold).