Simplify committing

[bimlas]

It's absolutely unnecessary to do anything with the commit message cleanup, Git will prepare and clean by self.

Previously when there was no COMMIT_EDITMSG exist (case of a newly created/cloned repository), the error handling try/finally dropped an error about the missing file, it's fixed now.

I hope it really fixes #37

[lambdalisue]

I'll investigate it after middle of March but quick questions

1. Is :e reload cache correctly? Users may close buffer without commit and opens again a bit later with extra changes. In this case the message part should NOT be changed but comment part

2. Do you have any source that said COMMIT_EDITMSG exsist for this kind of feature? I'm not sure if gina should modify the file or keep

[lambdalisue]

I've tested and the confirmed that this PR breaks a feature of caching.

1. Open Vim
2. :Gina commit --allow-empty
3. Write a commit message (say, This is a test message)
4. Save a cache with :w
5. Close the buffer without commit by :q and hit n<CR>
6. Execute :Gina commit again
7. A cached message should be shown but an initial commit message without saved cache is shown with this PR

So without fixing this breaks, I'm not going to approve this PR.

[bimlas]

I couldn't find a really better way to commit than the current solution.

What I tried:

• Read the contents of the current COMMIT_EDITMSG and store it in a variable (prev)
• Create a bare commit message to get the updated diff with git commit --no-template and store it (bare)
• Delete the lines of prev from the line matching the first line of bare, than append the contents of bare to it
• On BuffWrite write the contents back to COMMIT_EDITMSG, thus the saved commit message can be continued even after closing Vim and opening again

The diff is updated and the comments are recovered.

Problems:

• After commiting, Git does not removes COMMIT_EDITMSG, thus the content of it is never gets cleaned - I don't want to delete the file, because it would affetc Git (or other scripts) behaviour
• I could use some indicators (like === COMMITTED ===) on the last line, but this would be ugly

I think the current solution is better than my so closing the PR, but I don't like that it plays with reflog: it can be destructive in certain situation.