Merge pull request #899 from clojars/tobias/validation-improvements

Improve user validations

Don't typo check png files

[search] Rework search to always use db jar data

Search was originally pom-driven, then grew to support indexing either
pom data or db jar data. We only used the pom data when indexing on
upload, and ended up with a different :at timestamp value, depending on
the source of the data.

This simplifies the indexing to only support db jar data as input, which
fixes that discrepancy.

Fixes #897.

[api] Include desription in release-feed

Fixes #898.

[release-feed] Improve invalid from response

This gives a better indication of the correct format.

Merge pull request #896 from clojars/tobias/add-release-feed

[api] Add /api/release-feed

Fix invalid test macro

Adjust script size to 0.75GB

One goal of this memory tuning work was to _reduce_ the size of the
memory used by scripts, but in the process I went from 960MB (the
default Xmx on the current instance size) to 1024MB! This brings it down
to 3/4 of that to give us more headroom.

Increase script heap size

512MB was not enough for some of the scripts to work, so this doubles it
to 1GB. The instances have 3.75GB of memory, and the web app has a
1GB heap, so we can run two scripts alongside the web-app comfortably.

Constrain memory used by cron-run scripts

By default, the JVM will use a heap of 960GB on the current instance
size we are using if Xmx isn't provided. This ensures that the scripts
will use only 512GB, and will fully allocate it on startup to fail-fast
if it isn't available.

This also sets the jobs to exit if they OOM, and dump a heap in that
case.

This is to help prevent memory contention with the server process.