[phylo] fix dot output: render as plaintext, not JSON

See merge request !460

Fix allegedly-duplicated ngrams via better normalisation

Closes #526

See merge request !459

Phylomemy refactorings, to improve speed

See merge request !461

Resolve "Optimize Term table display"

See merge request !455

[ngrams] get rid of hardcoded "Nothing" string

See merge request !456

When building a social list, have MySelfFirst not silently override maplists

Closes #517

See merge request !454

This commit removes the `Int` argument from `cleanExtractedNgrams`,
because we were using it only in the `Flow.Extract` with a value of 255,
which meant that this function was essentially isomorphic to the
identity function unless we had very big compund ngrams (which feels not
realistic anyway). The other interesting branch of `cleanNgrams`, which
would call `text2ngrams` was never executed.

This changes the implementation of `buildNagramsLists` to fullfil the
requirement that David preferred, i.e. that merging and overriding are
really just a way to preserve or not the semantic structure of the
maplist, i.e. things like grouping or the ListType of each ngrams.

This reverts commit 5d6f1a5d.

What we were doing before to combine the existing terms wasn't correct,
it was just collaterally working due to the presence of the bug in the
combined ngrams that we fixed.

Currently the test for overwriting ngrams still doesn't work.

This commit extends the `haskellQuerySearch` set of documents used in
the `testMergeWithExisting` to nudge the `selectLems` function to pick
the "programs" postag and therefore join correctly later on.

I suspect there is a lurking bug somewhere as this seems quite fishy to
begin with, but hunting such but is outside the scope of this work, and
it will be done separately.

This commit gives a proper 'Semigroup' instance to `CommittedNgrams`
which has been now newtype-wrapped. Previously we were relying on the
stock deriving,which would piggyback on the left-biased combining
function for HashMaps.

This commit modifies 'buildNgramsLists' to fetch the existing terms
and combine them with the fresh (just extracted) ones.

This commit refactors the flow functions to stop passing a `Maybe
FlowSocialListWith` but rather just a `FlowSocialListWith`, where
functions that used to pass `Nothing` now just use `NoList`.

This also allowed us to get rid of some fragile states like in
`Gargantext.Core.Text.List.Social`, in the `flowSocialList` function
which had previously a `panicTrace` due to the fact the `NoList` could
still be passed to a function which by contract shouldn't have been
called with that.

This commit refactors the FlowSocialListWith type in order to:

* Split the "no list" vs the "yes list" cases into a sum type, so that
  we can pass to functions that ought to generate a list a type which is
  already a correct witness (without redundant information);
* Remove the boolean from the `NoList` constructor, as that conveyed the
  wrong information anyway.

This commit truncates dev output after 1000 lines (applies only to the
request and response middleware) so that we can get more screen estate
for what actually matters (i.e. debug logs).

When a corpus is delete, delete also the contexts associated with it

Closes #458

See merge request !446

Mostly guided by HLS

This commit fixes a bug in `purgeNodeCorpus` where we were calling
`deleteContext` on the nodeId of the node_context table entry, not the
actual contextId!

Expose via a test a subtle bug in `purgeNode` -- if the IDs of the
node_contexts and the contexts won't line up (as it would be the case
for the canned test data) the deletion won't work properly.

This commit fixes the helper I have introduced a few commits ago, which
was creating a new corpus for the user but it was returning to the
latter only the oldest created, not the newest.

That triggered a few bugs in the codebase.