gitlab_git 10.6.4 relies on Rugged marking blobs as binary or not,
instead of relying on Linguist. Linguist in turn would mark text blobs
as binary whenever they would contain byte sequences that could not be
encoded using UTF-8.

However, marking such blobs as binary is not correct. If one pushes a
Markdown document with invalid character sequences it's still a text
based Markdown document and not some random binary blob.

This commit overwrites Blob#data so it automatically converts text-based
content to UTF-8 (the encoding we use everywhere else) while taking care
of replacing any invalid sequences with the UTF-8 replacement character.
The data of binary blobs is left as-is.

This ensures that SVGs greater than 2 megabytes are not scrubbed and
rendered. This in turn prevents requests from timing out due to
reading/scrubbing large SVGs potentially taking a lot of time (and
memory). The use of 2 megabytes is completely arbitrary.

Fixes gitlab-org/gitlab-ce#1435

Signed-off-by: NRémy Coutable <remy@rymai.me>

This allows us to take advantage of Rails' `to_partial_path` to render
the correct partial based on the Blob type, rather than cluttering the
view with conditionals.

It also allows (and will allow in the future) better encapsulation for
Blob-related logic which makes sense for our Rails app but might not
make as much sense for the core `gitlab_git` library, such as detecting
if the blob is an SVG.