Django doesn't always generate SQL statements in a pre-defined order.

When adding or dropping multiple indexes, for example, it builds up

indexes as a set, and then iterates through them. This made

comparisons pretty unreliable.
We now pass the generated and expected SQL through a normalization

process. Sections of the expected SQL can be wrapped in a set, which

will then be sorted. The matching section of the generated SQL will be

sorted as well.
This is not perfect. If the generated SQL in that range is not a match,

it will still be sorted, and this can lead to some wonky results during

comparison. The "Generated SQL" output still uses the original generated

content, though, to help.
This also will not work quite right if Django ends up generating any

unordered SQL where each unordered item is actually a list of SQL

statements, but currently it doesn't appear that that's something that

happens. (Work was done to try to handle this case preemptively, but

it's far too complicated.)
Relevant series of statements has been updated to use this.

Unit tests pass for all supported versions of Python and Django.