mergeEditor: optimize removeDiffs from O(K*N) to single-pass O(N)

[xingsy97]

Problem

removeDiffs() processes each diff removal with two full-array passes (filter + map), creating 2K temporary arrays for K removals. Complexity is O(K * N) where N is the total diff count.

When accepting/rejecting all changes in a complex merge with 50+ conflicts, this becomes noticeable. Marked // TODO improve performance in the code.

Fix

Replace the per-removal loop with a single-pass algorithm:

1. Validate all diffs-to-remove exist using a Set for O(1) lookup
2. Apply all text model edits in reverse document order (same as before)
3. Single reverse pass through the diffs array: skip removed diffs, accumulate line delta, apply delta to remaining diffs

Complexity drops to O(N + K log K) with one array allocation instead of 2K.

[Copilot]

Pull request overview

This PR optimizes TextModelDiffs.removeDiffs() in the merge editor model to avoid repeated full-array passes when removing multiple diffs, aiming to improve responsiveness when accepting/rejecting many conflicts.

Changes:

• Replaces the per-diff filter + map loop with a single-pass rebuild of the diffs array.
• Adds a pre-validation step to ensure all diffs-to-remove exist before applying edits.
• Keeps bottom-up edit application to avoid range shifting during pushEditOperations.