This commits removes redundant access rights (ir.model.access), that are not really adding any extra rights when compared to those added to all internal users by the `calendar` module:

* xmlid: `calendar.access_calendar_event_all_employee`
* xmlid: `calendar.access_calendar_event_type_all`
* xmlid: `calendar.access_calendar_attendee_employee`

As this is just noise that adds no value, it's better to remove them in order to make it easier to maintain and customize them.


**Screenshots:**

In red, the removed Access Rights

<img width="1350" alt="Screenshot 2024-02-21 at 16 07 10" src="https://github.com/odoo/odoo/assets/1914185/29e40b85-eb91-405d-8120-495c5aa8942c">

<img width="1348" alt="Screenshot 2024-02-21 at 16 11 03 1" src="https://github.com/odoo/odoo/assets/1914185/11c743d1-82c7-4f3c-9c17-fa9251350cb0">

<img width="1348" alt="Screenshot 2024-02-21 at 16 11 55" src="https://github.com/odoo/odoo/assets/1914185/c6914a89-4b12-4acd-a61b-d4eb4ab3fbe0">


---
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

Add unit tests to prove a bug. No fix yet..

The added test is self explainatory, but here's some more context:

[EmailMessage](https://github.com/odoo/odoo/blob/c0ce5619ada6952261fd44f28108108040629d81/odoo/addons/test_new_api/models/test_new_api.py#L216) inherits from [Message](https://github.com/odoo/odoo/blob/c0ce5619ada6952261fd44f28108108040629d81/odoo/addons/test_new_api/models/test_new_api.py#L127) through its `message` field.



It's impossible to set a default value for this field, as it gets set back to `False` during the `onchange`.
It's even worse, the results are unpredictable. Related inherited fields will reference the value of the parent `message`, but the message itself will be `False`. (see `self.assertEqual(form.label, message.label)`)


I couldn't find a fix yet, but I tracked it down to this:

- https://github.com/odoo/odoo/blob/c0ce5619ada6952261fd44f28108108040629d81/odoo/models.py#L6415-L6418

Which makes the `record`, `snapshot0` and `snapshot1` to have a `NewId` value for `message` in the end, that gets converted to `False` during `convert_to_write`


--
I confirm I have signed the CLA and read the PR guidelines at www.odoo.com/submit-pr

`OrderedSet` inherited `-` from `collections.abc.Set`, which builds the
result via a generator and, when the other operand is itself a `Set`
(including another `OrderedSet`), tests membership through
`OrderedSet.__contains__` for every element - a Python-level call per
test instead of a raw dict lookup.

Specialize `__sub__`, mirroring the existing `__and__`: unwrap an
`OrderedSet` operand to its backing dict, use a plain `set`/`frozenset`
operand as-is, and only convert other iterables. This keeps membership
tests at dict/set speed in every case without adding a copy where one
isn't needed.

Benchmarking comparing the __sub__ operation of OrderedSet to OrderedSets
and on other iterables (Ordered set - othertype). The other object contains
1/2 of the objects in the original OrderedSet. The time values are averaged
across 200 runs.

| OrderedSet Size | Other Type | Old Time (µs) | New Time (µs) |
|---|---|---|---|
| 100 | OrderedSet | 6.8 | 4.6 |
| 100 | Set | 3.4 | 3.4 |
| 100 | FrozenSet | 3.7 | 3.4 |
| 100 | list | 8.4 | 4.3 |
| 10,000 | OrderedSet | 574.3 | 293.9 |
| 10,000 | Set | 288.0 | 274.6 |
| 10,000 | FrozenSet | 285.2 | 271.5 |
| 10,000 | list | 742.3 | 343.0 |
| 100,000 | OrderedSet | 5,851.7 | 2,968.0 |
| 100,000 | Set | 2,876.7  | 2,787.7  |
| 100,000 | FrozenSet | 2,864.3 | 2,787.7 |
| 100,000 | list | 7,692.2 | 3,678.5 |
| 1,000,000 | OrderedSet | 61,629.1 | 32,759.28 |
| 1,000,000 | Set | 31,377.5 | 30,295.9  |
| 1,000,000 | FrozenSet | 31,330.4 | 30,540.1 |
| 1,000,000 | list | 99,383.1 | 55.381.4 |