Drop custom CSV logger implementation

[carmocca]

The custom implementation in this repo has the issue fixed in Lightning-AI/pytorch-lightning#18567

Since we no longer keep the full metrics history in memory and write it every time, it's not straightforward to do the aggregation by step on the go.

This PR removes it, the csv file can be post-processed after training with something like:

import pandas as pd
df = pd.read_csv('metrics.csv')
combined_df = df.groupby('step').sum()
print(combined_df)