Better merge request manual instructions

I noticed the current instructions for manually merging a merge request that was submitted from a fork could create unnecessary merge requests or strange history. Imagine this:

1. Alice creates a repository and commits / pushes a single commit with SHA1 `A` to `master`.
2. Bob forks this repository and creates a new branch `myfeature` on `master` (`A`)
3. Alice commits `B` to `master` (which has the parent `A`)
4. Bob creates two commits on `myfeature` `P` and `Q`.
5. Bob submits the merge request to merge his `myfeature` into Alice's `master` branch.
6. Alice follows the manual merge request instructions:
  1. `git checkout -b bob/repo-myfeature master`
  2. `git pull http://... myfeature`

The branch `bob/repo-myfeature` was created from Alice's current `master`, which was `B`. When the `pull` is executed, git will fetch Bob's branch and then merged the fetched branch `Q` into the current branch's location `B`. This creates an unnecessary merge commit from `master` into the branch Alice is trying to merge. No harm is done, but the history is a bit messier.

This is even worse if Alice has set `git pull` to rebase by default. In this case, the commit `B` is rebased on top of `Q`. When Alice checks out `master` and merges in the branch, there will actually be a duplicate `B` commit.

These new instructions instead tell the user to fetch Bob's `myfeature` branch. This will fetch the necessary commits `P` and `Q` and create a temporary ref `FETCH_HEAD` pointing to `Q`. Alice will then create her local `bob/repo-myfeature` branch starting at `FETCH_HEAD`. No unnecessary merge commits, and no accidental rebasing.

See merge request !16