Inaccurate state setting in MJ simulation

## Type of Issue

Select the type of issue:
- [x] Bug report (to report a bug)
- [ ] Feature request (to request an additional feature)
- [ ] Tracker (I am just using this as a tracker)
- [ ] Refactor request
- [ ] Documentation Ask

## Description

While creating a mirroring simulation feature in Polymetis, a `set_state` method was added to all manipulator environments. To do so, the simulation is reset to the desired state / robot position and gravity compensation torques are added. A unit test was implemented to verify that the measured position and velocity matches the desired position and velocity. This unit test passes for the bullet env (within a tolerance of 1e-3), but fails for the MJ env.

## Current Behavior

Setting the position of a robot in MJ env using set_state yields a relatively good position but non-zero velocity and acceleration causing the position to grow increasingly inaccurate.

## Expected Behavior

Position and velocity should match with low error after setting state

## Steps to reproduce

The `test_mirror_env` test in `test_env.py` illustrates the issue.