When generating interdiffs, we figure out which chunks we actually want

to show any change information for, restricting it to chunks matching

the ranges indicated in the actual diff files.
The problem is, the way this interacts with "equals" computation (where

we strip two otherwise "equal" lines) can cause indentation information to

be filtered out. This happens because any chunk that isn't completely

within a valid range will get filtered out. Even though an indentation

change is an actual change line, we make it appear as an "equal", which

appears to span beyond those ranges.
To fix this, we needed to break up any chunks sitting on a valid range

boundary, keeping those parts of the chunks and turning the remains into

a filtered-equal chunk.
We then needed to convert the "filtered-equal" chunks to "equal" and

stitch those back together into "equal" chunks later. We already had

processors to do both of these things, but the one for merging adjacent

chunks happened right after interdiff filtering (which was too soon). We

now do this and convert back to "equal" in the post-processing meta

stage. At this stage, we also know about indentation information, so we

can also single those out as their own chunks, sanely.

Tested with the diff where I first noticed this. The indentation markers

were missing in an interdiff before, but are shown now. The lines of context

around it were correct.
Tested with my other interdiff review requests, and compared before/after.

They seemed fine. In particular, I tested with my regression test review request

designed to ensure indentation changes are hidden when outside of valid interdiff

filter ranges.
Unit tests pass.
We will want to pay attention in interdiffs after this goes in, in case there

are any regressions.