i18n for Admin Panel Settings section (WIP)

Add docs for perf monitoring a Metabase instance

Docs show the usage of VisualVM for local and remote (via JMX)
monitoring of a Metabase instance.

Fix async-test transient failures

Depending on the order in which the threads were running, we were
looking for the job in the running-jobs atom BEFORE we put the data in
the atom. If that happens, we might not ever update the job with the
correct status, it's also possible that we'll never remove the job
from the running-jobs atom, just depending on unlucky we are with the
ordering of the threads.

This commit introduces coordination between the code adding the new
data to the running-jobs atom and the work done by the job.

Fix typo in one insight string; clarify others a little

Fix more long lines

Change the metabase.middleware default log level to DEBUG

Switch alert-test `expect-with-temp` calls to `expect`

This will include API call timings by default.

Using `expect-with-temp` rather than `(expect ... (with-temp*` can
obscure some test failures. When using data from the `with-temp*` in
the left side of the `expect` it's needed, but if not, it's better to
nest he `with-temp*` so as to not interfere with the error handling
and test failure reporting.

This commit flips a bunch of the alerts tests that don't need `expect-with-temp`.

Add inline pulse rendering

This was broken from the last pulse related refactor. It affects slack
message attachments and the preview section of the pulse screen.

This introduces a render-type argument to some of the pulse rendering
functions where we can use the the correct behavior (bytes + base64
encoding or a URL + references) depending on how the image will be
used.

Testing this is difficult as this is creating images and then
depending on the render-type, encoding that image byte array into a
Base64 string. I added some very basic checks that at least ensures
the byte array is of the approximate right size. If we were rendering
it incorrectly, we'd see the byte array size much smaller. Pretty
fragile, but is the only way I could find to sanity check the results
so we don't introduce a bug like this in the future.

Infer spaces double to float

This change will consume about half the memory that the previous
doubles consumed. The change in precision shouldn't affect the results
of word splitting.

Support for Postgres enum columns

Code cleanup 3000

Add database_type to Field

Async test fix race condition, part IV

Export and pulse image attachment refactor

Previously images in pulse emails were serialized to bytes and the
bytes were kept in map stored in an atom. After all of the graphs were
serialized, they were written to disk and the file handle was passed
to the email library.

This commit immediately writes the file to disk and returns a
URL. This will no longer hold onto the memory need for all of the
cards and can be GC'd right away. This also avoids writing static
images like the no results image to disk evertime, rather than just
reading it out of the JAR.