Grid: Remove #initialChunkSize DOM pooling bypass & fix pool shrink bug

The Problem

1. The Shrink Bug: In #9474, the GridBody#createRowPool logic was updated to cap the pool size based on store.count. However, this aggressively destroyed DOM nodes when a Store was filtered (e.g. from 100 items down to 2), completely breaking the virtual scroller's "Fixed-DOM-Order" strategy which requires unused nodes to be recycled and hidden via display: none rather than destroyed.
2. The Chunking Bypass: When a data chunk was loaded (store.add(items, false) with initialChunkSize), the grid bypassed pooling and attempted to render the entire chunk into the DOM at once to provide "immediate scrollability".

The Solution

1. Prevent Pool Shrinkage: Updated createRowPool() to Math.min(windowSize, Math.max(me.items.length, countRecords)). The pool will grow to accommodate data (up to the viewport size limit), but it will never shrink during filtering operations, preserving existing DOM nodes for instant reuse when the filter is cleared.
2. Remove Chunking Bypass: Completely removed #initialChunkSize usage from GridBody. The grid now correctly utilizes standard row pooling (rendering only the viewport range) during a chunked load. Immediate scrollability is fully preserved because #initialTotalSize is still used to immediately size the scroll container bounds accurately.