debug/md

[Hsuchein]

Create a debug/md branch based on current enhance, and try to fix the bugs in current MD module, upload the visualized results to this channel after implementation and tests. Fix them one by one not in one shot. @AxiEj

1. High: The velocity-rescale thermostat formula is still suspect. The current enhance/dev branch retains this risk in vrescale.py (line 113). This was intentionally left untouched in the current merge round, as the conclusions for MD canonical sampling are not yet stable enough for release.

2. High: NPT still degrades to kinetic-only pressure when stress information is unavailable (utils.py, line 133). For common in-tree calculators such as MACE and AIMNet, this means "it runs" does not imply "physically acceptable."

3. Medium: Degrees-of-freedom bookkeeping is still hardcoded as 3N / 3N−3 logic and does not fully align with remove_com / remove_rotation settings (utils.py, line 68; vrescale.py, line 79).

4. Medium: On a fresh run the logger still backs up the current main .out file, which can split a single run's log across two files (logger.py, line 180).

[AxiEj]

In the new PR, I fixed problem 1.

"Fixed the bug in the V-rescale implementation and added conserved-energy monitoring to assess NVT sampling quality
(REF: Canonical sampling through velocity rescaling)"

Problem 3 solved in
"Fixed the missing support for COM removal and angular-motion removal (COM + rotation), and cleaned up the DOF logic accordingly"

Problem 4 solved in
"Fixed the bug where MD parameters were not printed to the .out file"

The rest of Problems are going to be solved within a week.