Set WM_NAME, WM_COMMAND, WM_CLASS on xsecurelock's windows for easier debugging and compositor configuration.

- Annotate some intentionally unused variables with (void).

- oclint found a bug in saver_multiplex that causes unnecessary screen
  saver restarts when receiving an XRandR event without a contentful
  configuration change.

This fixes:
- A potential bug in the "guessed monitor layout" code that may detect
  less monitors than correct (XRandR code path not affected).
- Unused values assigned to variables (annotated with a (void) cast
  where necessary).
- num_monitors variable redundantly checked for zero in cases where
  conditional compilation removes all code that would set it.

The configure macro had a small typo that made the default always check
- but luckily the Debian package isn't affected as it passes all desired
--with options anyway.

This prevents the input field from overflowing into neighboring
monitors.

This makes sure that the auth prompt always appears in the center of
every screen, and runs a separate screen saver on each one.

Further changes:

- Replaced an XSync by an XFlush, somewhat speeding up auth_pam_x11.
- Added a build dependency on XRandR to the Debian package.
- Global clang-format and IWYU.
- saver_xscreensaver now uses the parent process ID as random source
  instead of $RANDOM; this makes sure both monitors show the same screen
  saver.

This fixes cursor appearing on top of the screen lock in Cinnamon.

Revert "Explicit hide the cursor of the Composite Overlay Window." because this breaks the Cinnamon task bar.

This reverts commit e59e9575.

We seem to need this for Cinnamon but not for other desktops.

Fix a minor bug (uninitialized Pixmap data) in creating the empty cursor that might cause a black 1x1 point cursor.

The old code did not handle processes being terminated by signals right,
and basically waitpid() on them a second time, receiving ECHILD. This
didn't cause any real bugs, but was nevertheless incorrect.

This fixes artifacts that would come from deleting input after having
entered a password longer than the other text in the prompt.

We now require explicit opt-out to compile on systems without these being
installed; this prevents privacy issues from accidentally compiling with
XComposite support (which will let desktop notifications appear above
the screen lock).

Don't kill the saver_child's process group a second time when getting ECHILD and already having killed it.

This should not change anything noticeable - just a bit more consistent
that way.

saver_blank: Add configuration options for DPMS usage and blanking delay, and chain to it from all other savers.

Internal reference: b/73333712.

This pre-XRandR code guesses the number of screens by the resolution,
and may have guessed zero, causing an invisible (but working) auth
screen.

Add a log message when "Zaphod" configurations (multiple screens, not connected to a single desktop) are detected, as these are currently not implemented.

At least inside Xephyr, xsecurelock otherwise does not respond to
display size changes when the composite overlay window is not used (it
normally is).

Simulates multi monitor setups, but is currently quite limited in
functionality (also, resizing screens does not appear to work
correctly).

Internal reference: b/73107658.

Allow for up to three attempts on every PAM call unless a conversation error occurred (i.e. cancelled by user or timeout).

Also remove the EXIT_ON_CONVERSATION_ERROR and instead detect PAM
modules calling converse() a second time after having received an error
and exit then. Allows us to exit "cleanly" after conversation errors.

Also despam logging by hiding any errors that have been retried or
caused by a conversation error.

This gives the user one minute to retry logging in while not using
noticeably more power. Works around some displays that take really long
to wake up from "off" state.

This reduces the chance of burn-in due to misbehaving saver children,
and also allows improving saver_blank in a separate change by sleeping
before blanking (so one can quicker retry to log in).

Now the 1 second sleep after displaying an alert can be interrupt by a
key press; however this does not affect sleep times inside PAM modules.