subsys: fs: bm_zms: remove global callbacks array

[MirkoCovizzi]

Removes global callbacks array and replaces with callback
member inside struct bm_zms_fs.
This fixes a bug where registering multiple callbacks on
the same struct bm_zms_fs would update the user reference
number incrementally and saturate the global callbacks array.
When calling bm_zms_clear, this would only reset to NULL
the last callback set, leaving the rest still set.
This update makes it so that there can only be one callback
per each struct bm_zms_fs instance, mirroring the behavior
of other libraries.
Due to this change, also removes the BM_ZMS_MAX_USERS
Kconfig option.

[github-actions]

You can find the documentation preview for this PR here.

[rghaddab]

@MirkoCovizzi I do not understand this bug.
The current bm_zms behavior should be that only one callback is registered for a bm_zms_fs structure.
The caller will get its user_num which indicates its order in the callback register.
No other bm_zms_register should be called on the same structure before clearing the previous one.
Can you detail the usecases where this could happen ?

[lemrey]

I suppose that's one way to fix it. It does however add one limitation, which is that only the owner of an instance can receive callbacks from it, thus forcing a single owner for each instance. In practice, it is no longer possible to share a partition, every instance would need its own.

But it may be fine, first, because there a no events that are relevant for the whole application (like GC was). And second, because the application could still be designed in a way that allows to share a partition, by wrapping NVM operations.