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

Summary

In lib/core/database/sql_statement_splitter.dart:

1. Quadratic Line Counting ((N \cdot M)$):
   In close(int end) (lines 40–42):

   out.add(SqlStatementSpan(start: s, end: e, line: _lineOf(sql, s)));

   _lineOf(sql, s) (lines 147–153) iterates from 0 to offset counting newline characters (0x0A) on every statement. For a script with $ statements across $ lines, this performs (N \cdot M) \approx O(N^2)$ characters scans. In large SQL dumps or multi-statement migration scripts, this quadratic scanning takes seconds.

2. Per-Dollar Substring Allocations & Dynamic RegExp:
   In lines 67–76:

   if (c == '\$') {
     final match = RegExp(r'^\\$').matchAsPrefix(sql.substring(i));

   On every $ character encountered (including numbered parameters like $1, $2), sql.substring(i) copies the entire remainder of the script into a new string, and compiles a fresh RegExp instance. For a 5 MB file with dollar quotes or parameters, this creates massive transient allocations and GC pressure.

3. UI Thread Blocking:
   SqlStatementSplitter.spans is invoked synchronously on the UI thread in GenericSqlWorkspace._runSpans whenever the user executes a statement or script (Ctrl+Enter / F5).

Proposed Solution

1. Maintain a running currentLine counter during the main character iteration in SqlStatementSplitter.spans, turning line tracking into a strict (N)$ single pass.
2. Use a static compiled RegExp and pass the search offset start: i to matchAsPrefix(sql, i) instead of allocating sql.substring(i).
3. Offload statement splitting to an isolate via compute if script length exceeds a threshold (e.g. 64 KB).

Acceptance Criteria

• Statement splitting runs in linear (N)$ time with no quadratic line counting.
• No string allocations or dynamic regex creations occur on $ tokens.
• SqlStatementSpan.line calculations remain 100% accurate across all test scripts.