perf: cache the Redis folder key count; group aggregation parses numbers without copying

[ZhuchkaTriplesix]

From a data-structure review (claims checked against the code).

• RedisKeyFolderNode.totalKeyCount walked the whole subtree on every call, and the folder row calls it up to three times per rebuild (O(keys x depth) with nested open folders). It is now counted on first use and kept (late final); the engine hands out finished lists and does not change the tree afterwards.
• GridGroupingsEngine._computeAggregation copied every cell with replaceAll(',', ''). It now uses the comma-aware parser of the selection stats (parseThousandsSeparated, documented as equal to the copy); anything it cannot read takes the old path, so results are unchanged. The @visibleForTesting on that parser is dropped since lib code uses it now.
• Test: sums with thousands separators, padding, stray commas and a non-number.

[ZhuchkaTriplesix]

🔍 Post-Merge Code Review & Verification

LGTM! Excellent performance hardening across Redis folder trees and pivot calculations:

1. Redis Folder Key Caching (redis_key_tree_engine.dart):

   • late final int totalKeyCount = _countKeys() correctly eliminates the $O(\text{keys} \times \text{depth})$ recursive recalculation on every widget rebuild during tree scrolling.
   • Follow-up note: allKeys remains a dynamic recursive traversal. If batch operations on folders become frequent in future UX iterations, we can consider lazy-caching it similarly.

2. Zero-Copy Numeric Aggregation (grid_groupings_engine.dart):

   • Reusing GridSelectionCalcEngine.parseThousandsSeparated avoids intermediate String allocations during comma-separated numeric aggregation, while gracefully falling back to double.tryParse on edge cases.
   • Unit tests in data_grid_engines_test.dart cleanly verify comma-delimited sum calculations.