[3.x] Fix physics on_floor_body crash

[lawnjelly]

Physics body previously stored the RID of a collision object and accessed it on the next frame, leading to a crash if the object had been deleted. This PR stores the ObjectID in addition to the RID, and checks the object still exists prior to access.

Fixes #74732
Alternative to #77616

Notes

• This is far simpler conceptually to [3.x] Add RIDReferences #77616 and will be a more familiar pattern, and is thus easier to review (for non-core maintainers).
• The main disadvantage versus the proposed RID references is efficiency, ObjectDB::get_instance() used in this PR involved a lock and a hash table lookup, whereas RID references requires no lock and is very fast O(1).
• My own personal preference is usually to go for efficiency in areas such as physics, but I do understand if others prefer this simpler approach. I haven't profiled and it is likely in most games with small numbers of these objects there won't be a great difference between the two approaches in terms of performance.
• Like [3.x] Add RIDReferences #77616 this assumes a single threaded pattern. If another thread deletes the object between checking the ObjectDB and use, then all bets are off.

[lawnjelly]

@fabriceci would you be able to review this too?

It is essentially the same as #88947 in master.

[timothyqiu]

LGTM. I guess the same could be done to the 2D version as well?

[lawnjelly]

I guess the same could be done to the 2D version as well?

Absolutely, I must have missed this concentrating on the initial bug report, it looks identical. I might as well do in a separate PR though as this is already approved.

[lawnjelly]

Thanks!