perf(grid): eliminate dynamic RegExp compilations and substring allocations in GridFilterEngine #1527

Problem

In GridFilterEngine (lib/features/workspace/grid_filter_engine.dart), tokenize and _tryMatchPredicate are triggered on every keystroke in the grid filter bar.
On every step of the tokenization loop:

• input.substring(i) is called, allocating a new string.
• Over 9 RegExp instances are dynamically instantiated and compiled (AND|&&, OR|||, NOT|!, IS NULL, IN, BETWEEN, LIKE, comparisons, identifiers).
  For a moderate filter query, this generates ~50 temporary string allocations and ~60 dynamic regex compilations per keystroke, causing typing latency.

Solution

• Replace input.substring(i) with index-based checks (input.startsWith('AND', i) or codeUnitAt).
• Match comparison operators (>=, <=, !=, <>, ==, =, >, <) using branch logic or static table rather than dynamic regex.
• Precompile all keyword patterns into static final fields or replace with zero-allocation character scanners.

Benefits

• Zero regex compilations during keystroke tokenization.
• Instant feedback in grid search bar with negligible GC churn.