Update the tests and the implementation for retrieving historical data.
In the process, I simplified the response type. I think, in the future,
if we want to select unadjusted O - F or just the historical
observations, we should build that into the URL, rather than sending all
of the data back to the client and letting the client pick what to view.
This cuts down on data transfer and allows us to simplify our components
when they read data, since they don't have to parse a nested JavaScript
object.

Removed some of the dataclasses and functions that had been part of the
history API which are no longer used, now that we have the Parquet
implementation.

The diag.history function fails when there's a file:// prefix in the
diag file path because apparently Pandas gets a little confused and
tries to pass it to urllib instead of just opening the file.

This reverts commit 716a400.

Instead of deriving the Parquet file path from the Zarr path, the route
reads in the FLASK_DIAG_PARQUET variable from the environment and passes
that explicitly to diag.history. diag.history uses that to find the
Parquet file for this model, and strips out `file://` if it's present
because Pandas is a bit silly about that.

The charts really need to be refactored (in part) to eliminate these
hard-coded property accesses, but for now we can just update the
properties that the time series component expects so that we can include
test out the time series data.

Instead of deriving the path to save parquet fixtures for tests, we use
tmp_path, basically the same way we would for production.

It seems like pathlib.Path is having an issue with the `file://`
protocol on our environment variables, although I don't think it used
to. It's easy enough to strip this, although this may require more
robust path/uri handling in our tests.

When we process diag files in our pipeline, we convert the integers in
the diag files for is_used into booleans, so should treat them the same
way in our test fixtures. I think this is a sign that our test fixtures
are poorly set up, since they can get out of sync with reality. The
result of this problem was that either the application worked, or our
tests passed. Prior to this change, we needed to compare is_used to a
boolean for Parquet files generated with our ETL code, but to an integer
for test files generated with our fixtures.

…summaries