Pass lua_State* to inline destructors as well

[PhoenixWhitefire]

Currently, inline destructors set with lua_newuserdatadtor do not receive an L, while tag-based destructors do (#467).

This can be inconvenient if the lua_State* is needed during the destruction process, such as for a lua_getthreaddata, or to release a reference, where they are instead forced to store the L in their userdata object (e.g. @Bottersnike/eryx/src/modules/_fs_watch.cpp#L132). Embedders would be accepting the same risk that already exists with tag-based destructors.

This PR changes lua_newuserdatadtor to instead take a lua_Destructor, same as lua_setuserdatadtor, and passes the same L it would for tag-based destructors.

[vegorov-rbx]

We still think this is a risky direction to take, see the issue #1446 and discussion at #510

[vegorov-rbx]

We will probably be moving to take the path where we will try to api_check all invalid uses from inside GC (edit: in the future separately from this PR).
So with that and with some documentation about how bad it is too use 'L' that's provided, I think we can accept this.

[PhoenixWhitefire]

we will try to api_check all invalid uses from inside GC

Would this be like asserting the GC state in the barriers?

[vegorov-rbx]

Not asserting on a GC state, we will likely have a flag raised for the duration of the callback specifically.
It will be used for some new functionality coming in future weekly updates as well.