bug#56393: Actually fix the long lines display bug

[Gregory Heytings]

> Why is that a problem?  For this feature, we don't need the exact number 
> of modifications in character units.

Again, because I want to skip the long line detection code/overhead for 
normal editing operation (typing one character at a time).  I do not think 
that a 13 ms overhead for each typed character in a file with 150K lines 
is okay.

> Fine by me, but I think this is over-engineered.

Fine, but then you'd have to tell me how to do that in another way (that 
is, how to not trigger the long line detection code when the buffer has 
only changed a little), or to convince me that the 13 ms overhead is okay.

Also note that the full changeset only has 152 insertions and 88 
deletions, that is, 64 added lines.

> Also, please add a comment there explaining the heuristics, including 
> the cases we know about where the counters could behave "strangely" or 
> fail to catch changes.

I'll do that.