Shmessy

- New timestamp format: %Y-%m-%d %H:%M

- INTERFACE CHANGE - Replace the "ignore_virtual_types" by "types_to_ignore".
The new parameter get list of types to ignore

- New date format: `%d %b %y`
- Fallback to null improvements: apply the fallback after the column casting conversion (Performance improvement) + Change the fallback to np.NAN instead of python's None. This is due to the fast that some exporters does not supports python's None as a valid type (Export to delta/parquet for example)

- New date format with timezone: `%Y-%m-%d %H:%M:%S %Z`

- Fix issue when the column contains datetime.time data, the casting to unix_timestamp might raise an exception
- Identify data as float when the sampled data contains at least single non-empty data. This might fix the data casting for column with a lot of empty values that recognized as float instead of string

Hotfix: Allow casting to string from any source type. This fix will be helpful for mixed-type dataframes input such as dataframes that created from Excel files.