ZeroDivisionError on stock trades

**Describe the bug**
Divide by zero errors / warnings during model training.

`ZeroDivisionError: float divmod()`
`RuntimeWarning: divide by zero encountered in double_scalars`

**To Reproduce**
Steps to reproduce the behavior:
1. Run training (e.g. as in the notebook here: https://github.com/AI4Finance-Foundation/FinRL-Tutorials/blob/master/1-Introduction/Stock_NeurIPS2018_SB3.ipynb)
2. Will frequently report `RuntimeWarning: divide by zero encountered in double_scalars`
3. Will infrequently crash completely with `ZeroDivisionError: float divmod()` on the same line

Line in question: https://github.com/AI4Finance-Foundation/FinRL/blob/master/finrl/meta/env_stock_trading/env_stocktrading.py#L179

**Expected behavior**
Either the trade should be blocked, or the agent should receive a significant negative reward for making an illegal trade, rather than the environment crashing.