Enhancement: feedback and session comments

Currently, page feedback messages and session comments are hard coded in the database, that might bring result which are not satisfying, when l10n is enabled:

1. Some auto-feedbacks were translated (And, I've just found [current tests](https://github.com/inducer/relate/blob/83935e6770c747a6c42263038a53255d34b31d49/tests/test_utils.py#L263-L321) didn't cover that), for example

```python
ugettext_noop("Your answer is somewhat correct. (%.1f%%))" % (correctness * 100)
```
"Your answer is somewhat correct." will not have chances to be translated because it's saved into the db with the trailing percentage.

2. Saved code feedback/bulk_feedback displayed in the language of student browser.

For example, https://github.com/inducer/relate/blob/78af816a5b1bd69f51f3d336ec6621ebfb988d85/course/page/code.py#L1166-L1181 the feedback will always be displaying in the language of students' browser.

3. Mixed language in session comments(#575).

To solve this, we might need to introduce some models to save the META info of the feedback/bulk_feedback/flow_session comments, and use entry in those models to generate the rendered feedback/bulk_feedback/flow_session comments.