Draft: perf: Faster contact switching by caching spell checkers

This commit makes is snappier to switch between chat by caching the Gspell.Checker objects.

It triggers a linter error about a possible memory leak using lru_cache on methods but AFAIU it can be safely ignored since MessageInputTextView and TextBufferManager are singletons, see https://bugs.python.org/issue19859

Maybe 128 is too high a value for this cache, I'm not sure what's a sane size.

Benchmark

Procedure: switching between 2 contacts 20 times

Before patch

Name	Call Count	Time (ms)	Own time (ms)
_init_spell_checker	20	884	883
show_chat	20	1842	224
switch_contact	20	1088	185

After patch

Name	Call Count	Time (ms)	Own time (ms)
show_chat	20	823	223
switch_contact	20	3	0
_init_spell_checker	20	100	0