We have an increasing number of options that influence the display of

diffs. Most of these are site-wide, but some are influenced by user

settings (syntax highlighting in particular). Historically, while most

of these were queried during diff render, their values weren't factored

into any caching or ETags, requiring cache to be cleared for most to

take effect.
As new settings are coming for diff display, to govern code safety

checkers, it was time to rethink the approach here.
This change introduces reviewboard.diffviewer.settings.DiffSettings.

This is an option that represents all the settings that go into the

display of diffs generally. This includes things like whether syntax

highlighting is enabled, or how many diffs are on a page. It does not

include per-diff mode-specific settings, like whether to expand/collapse

a given section.
The DiffSettings object can be computed once and then passed around to

the various parts of the diff machinery. This keeps things consistent,

and avoids having to pass around individual bits of state or to

re-compute them.
It also contains a state_hash property, which returns a SHA256 of the

DiffSettings state. That's now included in all cache keys and ETags.

If any of these settings change, new cache keys and ETags are used. If

they're changed back, the original cache keys and ETags are used. This

avoids stale diffs and stays performant.
The old syntax highlighting flags for DiffChunkGenerator,

DiffRenderer, and populate_diff_chunks() have been deprecated, but

are still supported until Review Board 6. diff_settings is now

recommended, but each object will compute based on site defaults if not

provided.
This change intentionally does not add type hints for the functions that

were modified to take diff_settings, as adding type hints for these

functions is a surprisingly-large undertaking. I'll be revisiting those

functions in the future.

Unit tests pass on all supported versions of Python.
Manually tested this with the various diff viewer settings, including

some in-progress ones for code safety. Any time I changed a setting,

reloading would give me new diffs, verifying that both cache keys and

ETags had changed. Subsequent reloads with those settings used cached

state.