Break loops in Ngrams graphs

Closes #513

See merge request !453

Account for the fact we can now import ngrams terms containing loops.

components)

This commit rewrites the logic behind `buildForest` to build first a
graph, discovering loops in there (via the strongly-connected components
machinery) and finally converting the DAG (freshly-made so) into a
Forest.

This commit introduces the `OnLoopDetectedStrategy` type and does a
slight refactoring so that we can "plug in" algorithms to break loops
during our forest search.

It also introduces a test, which at the moment is failing, that we
can actually break a loop.

Resolve "On IMT Instance : Error message or crashes when changing the status of terms in large batch"

See merge request !445

Concurrent queries in NLP

See merge request !451

"[Server slowness] With the dev branch on the dev instance, we're experiencing a real slowness"

See merge request !430

Resolve "[Server slowness] With the dev branch on the dev instance, we're experiencing a real slowness" (JobInfo changes)

See merge request !429

This commit introduces the `OnLoopDetectedStrategy` type and does a
slight refactoring so that we can "plug in" algorithms to break loops
during our forest search.

It also introduces a test, which at the moment is failing, that we
can actually break a loop.

[The previous revert was a wrong track] Revert "Merge branch 'revert-97441f3d' into 'dev'"

See merge request !450

Also, implement pure `mapConcurrently` in extract ngrams, as it's
closely related to chunk size.

This reverts merge request !449

Revert "Merge branch 'adinapoli/issue-504' into 'dev'"

See merge request !449

This reverts merge request !439

Dev allow for api error in flow

See merge request !440

Dev worker fixes

See merge request !447

Dev add option to notify users

See merge request !438

Keep only the roots in searchTableNgrams

See merge request !439

[query] a prefixed query like ~prefix should transform to 'prefix', not '"prefix'

See merge request !443

This prevents multiple workers getting the same job at the moment it
times out.

In particular:

* In ngrams search tests where we were comparing the total number of
  results returned, we need to account only for the roots, because the
  total number we would be returning from the pagination would be
  higher, as it would include children, in some cases;
* In the `AcyclipTableMap` generator, the roots we generate need to be
  so (i.e. `_ne_root` and `_ne_parent` must be `Nothing`).

After we rewrote `destroyForest`, it means that in normal conditions we
are going to output roots followed by their nodes (in in-order
ordering), so the test outputs need adjustments.