Forest of trees: restore hierarchical grouping of terms

Closes #313

See merge request !424

Generalise fully 'GeneralisedNgramsTree' and add more commentary.

[sqlite] improve performance of sqlite generation by proper PRAGMA

See merge request !431

Also, force WAL checkpoint at the end, otherwise the DB might be incomplete.

This commit generalises the API we have seen for Ngrams and
NgramsElement to Ngrams and NgramsRepoElement, making it suitable for
the NodeStory manipulations.

Furthermore, we use zippers to efficiently traverse forests.

This paves the way for a more disciplined way of handling forests of
ngrams.

Revert "Merge branch 'adinapoli/bump-process' into 'dev'"

See merge request !427

This reverts merge request !426

Update process library to 1.6.26.1

See merge request !426

It turns out that we don't need to call `pruneForest` inside
`searchtableNgrams`. After all, we are just displaying information that
should be correct in the database from the get-go, so why we would need
any of that?

It turns out that the fix for #313 to support duplicated ngrams during
importing broke hierarchical grouping.

I have just temporarily commented out the code and I need to find a fix,
better also writing a test to ensure that grouping will still work in the
future.

This function allows us to keep only the roots, and discard the
redundant children after building a 'Forest'.

I think we can use the NgramsTree and NgramsForest types to our
advantage for the 'searchTableNgrams' & co, so I have revamped that
module to put it to good use.

This commit changes the implementation of filterNgramsNodes to not
filter too aggressively nodes that are present in a path.

This causes a test failure that needs to be investigated.

[openalex] make search act similar to the one on openalex web

See merge request !425

Port all the tasty specs to hspec

See merge request !423

See #490

Resolve "Error uploading zip file on dev"

See merge request !417

Fix bug in DB transaction rollbacks in the presence of domain-specific errors

Closes #480

See merge request !420

This commit fixes the bug with DbTx rollbacks and domain-specific
errors. What we do now is throwing an exception during the evaluation of
`DbFail`, to later catch it _after_ `withTransactionM` has performed a
clean rollback, and only at that point we call `throwError`.

This also means that we need to catch `DbTx` errors via the `MonadError`
interface, not via the exception-handling interface, to get proper
rollback behavior.

errorWith has a `HasNodeError` constraint but the name was misleading.

This means, of course, that now rollback tests are failing, but that's
fine, we can then fix things properly in the DbTx code.