CharacTER: MT metric (#286)

* init character MT metric

* Update README.md

* make style

* add isorts fixes

* make style

* fix example in README

* add cer dependency for tests

* Update metrics/character/requirements.txt

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Update metrics/character/README.md

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Update metrics/character/README.md

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Update metrics/character/character.py

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Update metrics/character/character.py

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Delete .gitattributes

* require cer >=1.1.0

* use calculate_cer when given a string

* add separate test for single/corpus

* streamline output format

the corpus version now only adds attributes, but cer_scores will always be present and always a list

* style

* update documentation

* add singleton example

* update cer dependency to 1.2.0

* make metric more robust

Now correctly accepts single strings and lists as input. Now only returns cer_scores and not other statistics as this seems rather uncommon and might be confusing for users.

* fix doctest formatting

* use non-local metric name

* update dependency

* simplify metric, assume we always work with batches

* aggregate scores

add aggregate and return_all_scores arguments

* add multi-reference option

* remove "Literal"

* Delete tests.py

Do tests via doctest instead

* Apply suggestions from code review

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>

* Update description

Co-authored-by: helen <31600291+mathemakitten@users.noreply.github.com>
Co-authored-by: Leandro von Werra <lvwerra@users.noreply.github.com>