Tracking: OverWindow performance optimization

[stdrc]

• Avoid rebuilding window frame information when handling every chunk.
  perf(over window): avoid recompute information about window frames #19036
• If changes in input row are not related to partition key, order key and window function arguments, the changes can be emitted to downstream directly.
  perf(over window): pass through Updates not affecting window function outputs #19056
• For window functions that supports incremental computation, jump to the first current key directly instead of sliding one by one.
• Make delta_btree_map::CursorWithDelta::move_next O(1) instead O(logn) now.
  perf(over window): O(1) impl for delta_btree_map::CursorWithDelta::move_next  #19214
• For row_number, rank and dense_rank, stop the iteration early when the output is not changed any more.
  perf(over window): skip remaining affected rows when rank is not changed #18950
• For dense_rank, Inserts won't affect "following" rows if they have the same value as existing ones.
• For rank and row_number, Updates won't affect "following" rows.
• For first_value and last_value, no need to iterate through the window.
  perf(over window): shortcut implementation for first_value and last_value in over window #19233