Rework the Discuss sidebar to match the new design. The tab navigation becomes a vertical icon rail on every viewport (it was a bottom bar on mobile), the "New" action and the search box share a first row, and the tab filters render as pill toggles on a second row.

Channel rows are restyled with rounded corners, an action-tinted hover/active state and an accent bar on the active row, replacing the global contrast() filter that read too dark on the light surface. The three surfaces now form a consistent elevation hierarchy in both light and dark mode: the rail is deepest (a faint pastel of the brand primary), the channel list sits above it, and the thread is lightest.

Subchannel rows abbreviate the parent channel to its first few letters so the subthread name stays the prominent label.

Channels without a custom image now get an avatar generated from the name initials over a stable pastel color derived from the record, instead of the fixed monochrome SVG.

task-6392371

https://github.com/odoo/odoo/pull/276874

The special social contribution (CSSS) is a quarterly tax, with quarterly
rates, that is paid monthly, with small advances.

The CSSS was only computed with the monthly rates, not using the correct
rates for couples with a spouse without revenue, and defaulting to
outdated rates if none were given.

I fixed it to use quarterly rates when needed (i.e.: at the last month
of the quarter, or for any payslip whose quarter already has a
validated/paid payslip at the end of the quarter), to get the correct
amount at the end of the quarter.

The payslip line was also missing in the 13th month, and termination
holiday (N and N-1) payslips. As they are subject to the CSSS.

If the employee does not respect fully his notice period, then his
IRC (which is the leftover salary he receives in his termination fees)
should be split per quarter left of the notice period, then its CSSS
should be computed as if the employee continued to work.

It was not possible to test the IRC from the UI, as scheduling a notice
period without respect would always reset
`l10n_be_notice_period_theoretical_end` to `False`. The bug was to fix
the computed spaghetti by splitting the `_compute_departure_date()` into
two different computes.

The parameter values for the were esoteric 2d/3d arrays that would
sometimes mix the employee status. Since, as per the POs, I only have to
worry about master for the Belgian Payroll, I have changed, split and
renamed the parameter values to better fit the use case.

DMFA tests are a pain to edit, and a hit to my mental health when
realizing that I have to rechange all the `expected_dict` by hand each
time I rebase. So I extracted the variable fields from the dict, and
made a function that helps us to easily compare the issue to avoid
wasting hours, or even days trying to adjust the tests.

I also added some variables in the dict, as some fields (e.g.:
`OccupationUserReference`) would change depending on the
employee/version id, which can change if the test is ran twice, as the
rollbacks do not rollback autoincrement indexes. So dmfa tests would
fail if we ran it more than once without dropping the db

The MISC ONSS account was removed from test_payslips_validation as the
payslip was in december (the M.ONSS was 0 euros as the total gross wage
was too low for the quarter)

For the account move line tests, the table comment seemed outdated (as
if the table was not modified but variables did). I deleted them as they
also seemed to be duplicate work to do.
Also, `_validate_move_lines()` would just iterate over a given array and
raise if the move lines would not be the same order as the given array.
`payslip.move_id.line_ids` does not seem to have a defined business
order, so I made it so that id does not fail if the recordset order is
different than the given array order

I also made two changes to the common localdict function:
- remove the `prev_` prefix to the keys, as it was wrong (the key can
also contain payslips in the future)
- Changed the logic so that it reuses yearly payslips when possible

task-6068542