Add support for CharFilter

[fulmicoton]

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

Sometimes, it is handy to have some text processing happen before tokenization, at the char level.
For instance, lowercasing could happen at that layer. Unicode normalization as well (#407).

Resolving HTML entities and removing tags could also happen here.
Handling Diacritics as well.

Describe the solution you'd like

Following lucene, we could have the concept of CharFilter that would happen before the tokenizer.
It might be time to rename the tokenizer module into analyzer or even "text", and have an analyzer
embed a charstream AND a tokenizer.

A bit of juggling will be required to keep track of the original char offsets.