Since Python 3.4, HTMLParser.unescape is deprecated and replaced with
html.unescape.

Since Python 3.3, some imports have moved from collections to
collections.abc but were kept for compatibility, though compatibility
will be dropped in Python 3.10.
We will keep compatibility ourselves for some time again and thus support
both imports.

Some documents are not ready to be downloaded, and when trying to download
them it results in a 500 error.
When now apply the same logic as in the site to decide if a document should
be listed or not.

Some documents may not be available yet, or may have been modified and are not
available.
Trying to download such documents results in a 500 error.
Thus we detect if the documents are available, using the same (javascript) logic
as the main site.

JS filenames are randomized and must be fetched. One of the random
strings has been moved from a <script> tag to another js file that we
need to fetch.

This can happen because it is displayed twice (bank statement + deffered
statements). But the attributes of the subscription were the same.
Before this changes, the duplicates were supposed to be handled by the
variable parsed_subscription_ids but its values were given only at the
first iteration, so it was just an empty list and the condition was
always True.
Since the regexp which get some information on the subscription is now
used once per subscription, there were no need to make a method of it.

Sometimes the details page is not yet accessible for some accounts.
Prevent crashing in this case.

* price and vat are not always available
* day at first in date
* handle pre_taxe_price

"possible" instead of "disponible" in the xpath. We keep handling both
messages.

When this authentification method is the default choice, we check if we
can use PASSWORD authentification mode.

The card index was hardcoded to "0" so we were always fetching the same
transactions for all cards.
Also handled balance for cards that don't have an "en cours" balance.
Also handled the "no transaction" case.

"No captcha" was the nickname of RecaptchaV2 service, but it's confusing
and less precise. Rename and keep compatibility.

Authentication seems to have changed (#453), and in fact a few changes are needed
to make it work again:
* A new js page is now the target of the initial redirection chain,
* This page gives us a new parameter ('goto') which is needed by the authenticate
call
* and old cookies need to be cleared before the authenticate call (otherwise an error
is thrown)

In Value.check_valid method, regexp in validation error message
is not consistent with actual matching test: it uses re.match
and adds ending anchor.
We can "normalize" regexp in objet initialization to be consistent.

A logger signaled that additionnal transactions where no longer loaded,
compared to the minimal list shown when reaching OperationsPage the
first time.

Now get_form has been updated.

Sometimes, on creditfoncier, the wrongpass message is raised as a websiteUnavailable error.

Only for "professionnels" connections (current BProBrowser). It is the
first part of the rework, other changes will come next. What is
addressed so far:

- login without sca, accounts, history, profile

By the way, it also has the benefit of removing dead code: the variable
`accounts_and_loans_url` doesn't seem to be used

TODO:

- implement the SCA

- check if there are changes for wealth and bill with appropriate
connections, and do the changes if needed

- add new account types, etc...

The pro website has changed, and an url '.*voscomptes/identification/identification.ea.*'
that was once handled as a temporary unvailability page is now an
expected page of the login process.

I don't know if this url has to be treated the same way or differently for
the par website, so I decided to override the parent attribute in the
subclass BProBrowser.

We can also notice the following requests don't work because the pro website
has changed (a wrong request will typically lead to a deconnection, for
ex. the next_request auth_page() fails this way), but not for all
connections (strange), meaning for some connections, the current
requests work fine. So far I have noticed the current code still works
for connections from users who have already validated the SCA (but I
might have missed something else).

Allows to match the origin account by iban instead of by id.

The portfolio page takes more than 30s to load.

The name of a query parameter has changed, so the redirection URL was
not recognized.

Some invests were missing because we only fetched the first `<div>` table.
This works well with aviva but for the abstract module "afer" we must
filter the `<tr>` from the "Répartition" table by counting their `<td>`.
Also handled another xpath for opening_date.