[3.x] Move Bullet physics query flush from Bullet space pre-tick callback to Bullet physics flush_queries() again.

[madmiraal]

This reverts commit 10544f1.

As per discussion here, #40185 was reverted for 3.2.3, because of a regression issues. Once all the regression issues have been resolved this change can be reapplied.

Regression issues that need to be resolved before applying this PR:

• Setting RigidBody global_transform into the _input function fails. #40486
• [Bullet] Gravity is applied to RigidBody even with custom_integrator turned on #40508 - Fixed with c31acd0 or a 3.2 cherry-pick of Do not set gravity and damping when Custom Integrator is enabled. #42650.
• Removing a CollisionShape or CollisionPolygon from a sleeping RigidBody at runtime breaks gravity in Bullet physics #40739 - Fixed with [3.2] Reload Bullet space override modifier even when RigidBody is inactive. #40990

Edit: I've added the fix for #40508 as a separate commit to this PR.

Closes #37702

[akien-mga]

I think re-cherrypicking the fix might look better than a revert of a revert (which looks like we messed up twice :P), but from a purely technical point of view this is fine :)

[madmiraal]

Reverting the revert was the easiest way to ensure it was correct. Cherry-picking from master fails, because of the file renames and would require manual changes. Ultimately, it's just an automatic commit message, which can be changed. If you want me to change it, let me know.

[madmiraal]

Since #40252 also has a regression (#40840), I've added the fix for #40508 as a new commit to this PR.

[jknightdoeswork]

Necroing this issue. Just discovered that the Bullet Physics system is 'off by 1' frame because of this. ie: body.position is off by 1 frame from what is in the physics engine. This should probably be merged. On my branch, I just tried moving the flush from the 'pre tick' to the 'post tick'. Seems to work great.

Would like @madmiraal or @AndreaCatania to comment here as it seems like there were some issues here.

Nevermind, it's still of by 1. The problem is even deeper. The "godot motion state" used by the bullet server is introducing some kind of physics interpolation at a deep level. I'm disabling that completely. For clarity: I'm working on networking the bullet physics server. I have it working but I have to use PhysicsServer::get_body_state() in order to get the real position of the object, rigid_body.translation is unusable and contains invalid values.

I am capable of networking the Godot physics server in Godot 3 as well and will be doing that soon.

I can create a PR with the changes I've made to the Bullet physics server to enable it to be networked, but it would only be worth doing if some of the frequent contributors like madmiraal or Andrea Catania want to review it. It seems like the Bullet physics server has a lot of 'edge cases' and band-aid code over various systems like the kinematic character controller, etc, that I'm not very familliar with and likely will not heavily test.