Expand Verifier Type to Add Support for Multiple Heuristics

**Is your feature request related to a problem? Please describe.**

Currently, every verifier type has a single implementation. For example, `math` verifier doesn't offer flexibility on how strict we would want the reward function to be regarding the formatting of the output.

Instead, it would be great if the strictness of the verifier could be configurable.

**Describe the solution you'd like**

[This notebook](https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Gemma3_(1B)-GRPO.ipynb) from unsloth shows an example of how a simple implement for this could look like:

**Additional context**

Add any other context or screenshots about the feature request here