bug#64391: buffer narrowing slowdown regression in emacs 29

[Mattias Engdegård]

How expensive markers can be when a buffer is modified is often unappreciated. 
Here are the results of a benchmark to illustrate the cost (code attached).
A buffer contains "abababab...", we optionally add a marker to every other 
position, and then time the deletion of all the "a"s. Results:

|   size | without markers | with markers |
|--------+-----------------+--------------|
|    500 |           0.000 |        0.001 |
|   1000 |           0.001 |        0.004 |
|   5000 |           0.002 |        0.089 |
|  10000 |           0.005 |        0.363 |
|  50000 |           0.024 |        9.527 |
| 100000 |           0.049 |       58.048 |

Now there's something for the curve-fitters among you.

abench4.el
Description: Binary data