Allow the datasource to decide how to handle a
canceled request.

Responsibility is now to pass the cacheKey through
to `datasourceRequest` in each datasources'
implementation of `query`.

Repeat Prometheus requests with the same query
will cancel preceding requests.

Repeat requests, if they already exist in the
in-flight request map, will cause the previous
request to cancel.

The implementation of the unique key is the
responsibility of individual datasources.

Enable the "limit" param in /api/search

Conflicts:
	CHANGELOG.md

Commit adds a check whether Organization exists before User is added to
the organization.

Fixes #3151.

restart grafana after upgrade

I've forgotten to update the systemctl start to restart in my first commit

The current script only tries to start the already running grafana process after an upgrade. This leads to errors due to changed js and css hashes while grafana is delivering the old ones. To load the new sources we need to restart grafana after an upgrade.