This was useful in the past as parts of the Exposure API were still in
flux, but it's stable enough now that this shouldn't be necessary.

This snake_case stuck out, so I fixed it.

* Rework loop to simplify logic.
* Remove "catch all exceptions" logic, replacing with just catching
AstrometryError (which could be raised by the fitter or matcher).
* Remove "using previous iteration" which appears to be a hack added
on DM-2755 to deal with problems they were having with the old matcher.
* Remove/rename variables to hopefully reduce excessive indirection.
* Expand tests to reflect the new exception logic.
* Remove testWcsFailure that was based on "never raise" approach.

* Remove testUsedFlag and incorporate its test of calib_astrometry_used into
the all the `doTest` calls.
* Removed unneeded tearDown.
* Use more standard `task` instead of `solver`.

It is not used elsewhere, so no need for it until we're done.

Previously, AstrometryTask would pass if it just kept the WCS on the
exposure the same: the test now starts with a "wrong" (undistorted) WCS
and will fail if the output WCS matches the input (at least for the
testRadial case).