bug#70077: An easier way to track buffer changes

[Stefan Monnier]

> Ah, yes, you are correct, I had missed that one. As you note, it would
> be costly, especially because if you wanted to do anything with that
> data, you would probably end up dumping it into a temp buffer.

That's indeed what I end up doing in the `eglot.el` case.
But if the API wants to tackle the performance issue of combining many
small changes into a larger one, I don't know how we can do better.

Another idea I considered was to keep a whole buffer as "previous
contents" (instead of a string that covers the modified area).
In some cases it would be more efficient, but the common case would be
a lot more costly.


        Stefan