* Add "view on site" links in Check, Channel, and Project admins
* Expose over_limit_date in Profile admin
* Display last notify duration ("Time" column) in Channel admin
* Add last notify duration filter for Channel admin
* is_past_over_limit_grace(): Returns True if this profile is
over limits for 31 or more days.
* schedule_for_deletion(): Sets the deletion_scheduled_date
field to 31 days in the future.
In the "Details" and "Log" pages Healthchecks displays a list
of events (incoming pings and sent alerts). At the top of the
events list is a two- or three-way selector for selecting
the timezone for formatting event dates and times. The selector
options are "UTC", check's configured timezone, and "Browser's
time zone". The "Browser's time zone" used to be default, initial
selection for all checks.
With this change, for checks that use cron schedule, the default
selected timezone will be the check's configured timezone.
The "Browser's time zone" option is of course still there and the
user can switch to it to see dates and times in their local time.
Rationale: I semi-regularly get support requests about unexpected
or missing alerts, where the problem boil downs to a timezone
mismatch between the client and the Healthchecks server. Sometimes
the confusion seems to be caused by the user seeing ping arrival
times in their local time zone, comparing them to their cron
expression, and not realizing their server may be using a different
timezone. By switching the default display timezone to the check's
configured timezone, I hope users will be more likely to notice
discrepancies between ping arrival times, the cron schedule,
and their local clock.
For checks using simple schedules (timeout and grace), we still
default to browser's timezone for display.
Fixed CSS issues:
* In the tag input element, when entering a new tag,
the "Add ..." item was missing padding. Padding added.
* In the timezone selects the caret was hidden, so it was
hard to discover that timezone options can be filtered by typing.
Caret in timezone selects is now visible and blinking.
The smtpd management command runs a SMTP listener for receiving
ping signals as emails. If emails arrive infrequently, the
database connections can time out. A brute-force workaround for
this was to call `connections.close_all()` before making new
DB queries.
Code moved around while migrating to aiosmtpd, and looks like
the workaround did not work any more – #847. This commit
replaces `connections.close_all()` with `connection.close()` and
moves it to the `_process_message()` function, which *hopefully* fixes
the problem.
