Don't create a `merge_bases` entry in the database if no merge base
argument was passed to the script. The `get_merge_base` function in
`code-review-database` will default to `origin/master` or `master` when
no row is found in the database. Rely on those defaults instead of
always inserting a row.

Put these functions in a function-only file that can be sourced by other
scripts that need database access.

Function to get the merge base for the current branch. If none exists in
the database, default to `origin/master` if there's an `origin` remote,
or `master` if not.

Thanks to Serpiton
(https://codereview.stackexchange.com/users/61525/serpiton) on Stack
Overflow for describing how to get a default value using the `MAX()`
function:
https://codereview.stackexchange.com/questions/74192/select-first-value-if-exists-otherwise-select-default-value/74252#74252

Thanks to 'tig' (https://stackoverflow.com/users/96823/tig) on Stack
Overflow for the command to check if a given Git remote exists:
https://stackoverflow.com/questions/12170459/check-if-git-remote-exists-before-first-push/26843249#26843249

This command will store the desired merge base for the current branch,
defaulting to "origin/master". In so doing, subsequent `code-review`
subcommands will have access to this and be able to default to a merge
base branch instead of:

* Having users specify one manually at the command line for each
  `code-review` subcommand
* Having a default merge base branch that's not desired by users

I figure I can hopefully get away with not sanitizing SQL input because
I don't think you can use apostrophes in Git branch names. Otherwise
I'll probably have to port this to another language with a SQLite
library, as it doesn't seem possible to easily sanitize inputs in Bash.