We do not need to read the same raws in every ingest check.
We also do not need to read every raw dataset that was ingested.
Read the first raw completely but for subsequent raws
just read some metadata to demonstrate that the datastore
can read them.

Changing this can make significant improvements to test
runtimes.

It doesn't depend on any of the loop variables.

Previously the in place code was never triggering and it
was picking up the original file. Now explicitly override
the file being ingested and check that the URI matches our
expectations.

Register instrument takes a while for some instruments
and it always does the same thing and can be reused for
each test.