Also, changed exceptions to safe

- Re-enable orphan warnings that were manually disabled in many modules
- Remove non-Haskell files that were just sitting in the source tree
- Remove modules that were not called from anywhere
- Remove unused dependencies

This is not exhaustive by any means. In particular, some more weeding out
can be achieved by looking at individual functions and even branches of
functions. Weeder can help with that.

Also, fix for notifications throttle delay value

Also, throttle messages to avoid flooding the client

It forces programmers to think about errors we are logging and reporting
to the frontend, because they need to contain no sensitive data.

WARNING: This sets the last mod time as UTC. Issue filed here for
explanation:
https://github.com/jgm/zip-archive/issues/67

Also, some test refactoring and add servant-client to tests.

Removing the `All` data constructor from  the `Lang` datatype has
several benifits, the main one being that we now have total functions
for things like `toISO693`.

This will make possible to write an uniform interface for stemming
`stem :: Lang -> StemmingAlgorithm -> T.Text -> T.Text`, as now
`Lang` will always map to a valid (ISO693) language.

It's a useful function to log (with `DEBUG` severity) the execution time
of various functions. Debug logs won't be printed in production.

We do that thanks to a new typeclass method for
the `MonadJobStatus` called `noJobHandle`.

Basically, Gargantext.Prelude exports all of Protolude now.

This fixes commit 5d5300cd which
automatically changed all 'return' to 'pure' in our code, resulting in
broken SQL etc.

Fixes #258

This commit fixes the issue of the "empty Ngrams". The culprit was a bug
in the `arxiv-api` package, which was fixed by:

gargantext/crawlers/arxiv-api!2 (commits)

In a nutshell, due to the fact the relevant `Conduit` was never
completing, it was blocking the `flow` function, which was endlessly
waiting for the last part of the results to arrive.

Hence we were never calling the `flowCorpusUser`, which is the function
responsible for generating the Ngrams.

This commit moves the `threadDelay` present in the `gcThread` function
at the beginning of the `go` aux rather than at the end. This has the
effect that the first GC happens _after_ the delay time and not
immediately, which might make some tests unpredictable if we get unlucky
and some job ids are removed from the underlying map before we can
inspect their state.

We need to fix them properly as part of #198.

This reverts merge request !166

JSON should now be better so that the UI can handle it in task async
handler.

JSON should now be better so that the UI can handle it in task async
handler.