Fix stalled HTTP fetches with large payloads

Closes #92, gitlab-ce#25916, and gitlab-com/infrastructure#941

See merge request !110

Add helper.IsContentType and use it everywhere

Closes #90

See merge request !114

Add unit test for upload-pack handling

See merge request !111

Allow nested namespaces in git URLs

Closes #74

See merge request !80

For fetches over HTTP, Workhorse executes git-upload-pack and first
attempts to send all the input data to stdin before reading from the
stdout pipe. However, when the payload is large, the stdout pipe may
fill up, causing git-upload-pack to stop reading from stdin. Workhorse
will then be deadlocked, since it will be waiting to send more data
to a buffer that will never be drained.

An addition side effect is that git-upload-pack processes also get left
around. These processes are cleaned up only after Workhorse is
restarted.

This fix modifies the git-upload-pack behavior to consume the entire
HTTP input first so that reading the data from stdout and sending the
reply can be performed in a separate Goroutine.

Closes #92

Closes gitlab-org/gitlab-ce#25916

Closes gitlab-com/infrastructure#941

Proxy GET /info/refs to Gitaly

See merge request !105

Catch _all_ multipart NextPart() errors

Closes #89

See merge request !108

Replace 'gitlab_workhorse_artifacts_upload_*' with labeled version of 'gitlab_workhorse_multipart_upload_*'

Closes #88

See merge request !106

Set correct value of X-Forwarded-For header in PreAuthorize request

Closes #82

See merge request !104

Replace 'gitlab_workhorse_artifacts_upload_*' with labeled version of 'gitlab_workhorse_multipart_upload_*'

Version 1.2.1 changelog

See merge request !103

Add CI related requests metrics

This MR adds some metrics for CI related requests.

Related to gitlab-org/gitlab-ce/#23366, gitlab-com/infrastructure#583

See merge request !97

Prometheus metrics for multipart file extraction

See merge request !100

API improvements: buffering, metrics

Following discussion in https://gitlab.com/gitlab-org/gitlab-workhorse/merge_requests/94

This closes a hole in our Unicorn response buffering. The responses should be small, hence buffering in-memory.

See merge request !102