Before this commit, during a module uninstall, module data would be
unlinked on a record-by-record basis, this was done this way for two
elementary reasons:
    

-  To guarantee an unlink order, as some module data being deleted
    might only be deleted if some other module data has already been
    deleted.
        e.g. [a, b, c, a] -> we could batch unlink both `a` in the
             list, however the second `a` could depend on the unlink
             of either b or c, making the groupby erroneous.
-  To create a savepoint before every module data unlink (record), so
    that in case it fails, we can simply skip it and proceed with the
    uninstall.

After this commit, batch unlink of module data is now possible without
losing the aforementioned key features.

- The first problem is fixed by using `itertools.groupby`, which
    will group **adjacent** data of the same type and unlinks them all
    in batch while guaranteeing correct unlink order.
        e.g. [a, a, a, b, b, c, a, c, c, c] -> this will unlink in
        batches of [a, a, a], [b, b], [c], [a], [c, c, c]

- The second problem is fixed by making a savepoint per batch, if
    a single element of the batch fails and the recordset contains more
    than one element, we split the recordset into two and we repeat the
    unlink process recursively until we weed out the undeletable record

Real world example: **l10n_mx_edi** takes **20min** to uninstall because it
contains a lot of master data (~55k records), with this patch the
uninstall takes a measly **20s**.

It should generally reduce the uninstallation time for all modules, but
those with lots of master data will see greater benefits in terms of
time.

opw-2025895

cc @Julien00859