[FW] Bunch of fixes for Binary/Image fields

[fw-bot]

[FIX] core: add invalidation of Environment's _cache_key.

Changing the environment in method create() to force bin_size=False
looks harmless, but it actually breaks many tests, in particular in
module account. The reason is that company_dependent fields are read at
the wrong place in the cache. And this is because env._cache_key can
be polluted with old data.

Make sure that _cache_key is cleared when resetting all the lazy
properties on the environment. Only the change in res_user.py makes it
work, but let's not tempt the devil.

Side note: I hate caches.

[FIX] test_new_api: test's change that already works

Add some test cases, and also remove useless flush/invalidate.

[FIX] core: fix create/write on binary fields

When invoking create() or write() with a binary field, the cache of the
field was incorrect if bin_size=True was in context. Force context with
bin_size=False when putting a binary value in cache. It is particularly
important to have coherent values in the cache for web_save.

Also, because an environment with bin_size=False won't return the same
context cache key as one with bin_size=None, it leads to have a cache
inconstistency when we write with bin_size=False. Change Environment
method cache_key() to return the same cache key when bin_size is absent,
bin_size=None or bin_size=False.

Tests on binary fields have been updated to not rely on flush and
invalidate. We also created specific tests for the binary write() path.

[FIX] core: cache inconsistency when assigning related resized image field

After writing or creating on a related Image field, its cache contains
the full-size image instead of the resized one (according to its
attributes max_width and max_height). Fix it by re-setting the
the resized image on the cache at the end of the inverse method.

[FIX] core: flush non-attachment binary field when necessary

Non-attachment binary field needs to be flushed before reading their size,
since the latter relies on the database's binary size function.

[FIX] core: fix no-attachment binary for web_save

When bin_size=True is in the context, the computed no-attachment binary
was incorrectly saved to the database. The row is updated with the size
of the binary instead of the value itself (check compute_value() +
cache.set).

Fix this problem by avoiding setting the cache of the bin_size value
as dirty.

Moreover, the binary size is computed with pg_size_pretty for the
no-attachment binary fields. Also, in compute_value() we called
b64decode on the value that was previously encoded in base64 by
_compute_datas(). But in our case, we never use _compute_datas()
because it is not stored as an attachment. Then b64decode() doesn't
make sense in this case.

These 2 bugs are now covered by testing web_save(), where cache
consistency is required. It was first reported for this method.

First proposal: #156673

Forward-Port-Of: #163634
Forward-Port-Of: #160708

[robodoo]

Pull request status dashboard.

[fw-bot]

This PR targets saas-17.2 and is part of the forward-port chain. Further PRs will be created up to master.

More info at https://github.com/odoo/odoo/wiki/Mergebot#forward-port