Each static type is uniq and global for all subinterpreters and can't
access cached PyObjects. Decoder do use cached PyObjecs and must access
cache stored in module rather than in static variables.

Static type it can't access "proper module" due to its "per process
singleton nature". The recomended way is to turn static types into heap
types in order to use cached PyObjects.

See
#227 (review)
for an explanation.

This is a first step in the direction of subinterpreter compatibility.
Is's recommended to keep such caches in the module state rather then
process wide globals:

https://docs.python.org/3/howto/isolating-extensions.html#managing-per-module-state

Dedicated struct was created which can later on be stored in the module
state. It can't be done right now since those members are accessed from
static types initialization code which doesn't have access to the module.
Turning such static types into heap types is also subinterpreter
compatibility prerequisite:
https://docs.python.org/3/howto/isolating-extensions.html#heap-types

I'm not really sure this is required, because AFAIU all the API is
actually static, and apparently shared by all interpreters... Anyway,
this seems clearer.

Both are heap types now: their instances hold a reference to the type and,
being GC-tracked, must be untracked before being freed. Their dealloc was
only calling tp_free(), leaking one reference to the type per instance and,
on debug interpreters, emitting a ResourceWarning.

Python 3.10 is EOL since 2026-10-01, and its not easy for me to even
test for it locally.

This also reverts commit 82b1453.

Its configuration still referenced Python 3.9 and 3.10, a good evidence
that it was abandoned lot of time ago.