This "conversion" process now has two levels:
The low-level part is in `valuetype`, where ics value strings are converted to
simple python objects and back, based on some externally determined VALUE type
and without using any context or inspecting parameters.
The high-level part is in `converter`. A `GenericConverter` can take multiple
ContentLines or Containers and convert them into the respective values of a
`Component`. For `AttributeConverter`s this is done for a single attribute of a
`Component` type (which must be an `attr`s class), using metadata of the
attribute to determine further information, such as value type, is required or
multi value, ics name, etc. `AttributeValueConverter`s then use a `valuetype`
converter to parse the simple attribute value. Other converters combine
multiple `ContentLine`s into a single attribute, e.g. a `Timespan`, `Person`,
or `rrule`. The `ComponentConverter`, which is created from the `Meta`
attribute set on any `Component` subclass, inspects all attributes of the class
and calls the respective converters. All unknown parameters are now also
collected in a dict.

This makes it a lot less work to add new attributes, as most conversion logic
can be generated automatically and without any redundant code. Additionally,
this makes it easier to implement correct handling of `VTIMEZONE`s in all
places and will allow implementation of JSON-based ical handling and variable
levels of parser strictness.

I also included some further refactorings: the `tools` module with the broken
online validation is gone (the website is offline), as all `Alarm`s now only
take a few lines they have been merged into a single file, `ics.grammar.parse`
has been shortened to `ics.grammar`, the inner `Meta` classes has been replaced
by instances of an `attr`s class, the `Component` conversion methods are now
called `from_container` and `to_container` and for `ContentLine`/`Container`
there's now a `serialize` method to convert them to ics strings. The most
important change might be that all custom `__str__` and `__repr__` methods were
removed. They now default to what `attr` generates and in general follow the
standard that `repr`s "should look like a valid Python expression that could be
used to recreate an object with the same value". This makes debugging easier
and allows us to implement `str` with a nice and short informal representation,
and only generate the ics representation when this is intended. Previously,
`__str__` returned the ics string and `__repr__` returned a very short informal
description, which made dumping the actual python values hard when debugging.

ValueConverters are now allowed to modify optional params and context, i.e. consume params and store context when parsing, and add params when serializing.
also move ExtraParams to types, use NewType instead of a direct alias to catch invalid dict usage, and ensure that they are copied using deep-copy (they might contain lists),
add EmptyDict as argument default, fix timespan context clean-up

see python-poetry/poetry#1941 (comment)

this was done using `bumpversion --verbose release`

this was done using `bumpversion --verbose minor`

this was done using `bumpversion --verbose release`

…st tests/) to fix PATH problems

`noqa` and `type: ignore` are now only used for actual bugs in the checkers
unfortunately, current pyflakes dislikes `type: ignore[something]`, so we can't ignore specific mypy bugs until pyflakes 2.2 is in flakes8

Co-Authored-By: Tom Schraitle <tomschr@users.noreply.github.com>

Mostly by moving/splitting test dependencies to different sections in
tox.ini as mypy and pypy don't work well together and it is sufficient
to run mypy checks on CPython.