In the thread documentation, I noticed that unsafe signals preclude
killing threads with a signal.  But this is the normal method for
clearing a watchdog timer under threads.  I doubt that people are
compiling with PERL_OLD_SIGNALS these days, so I didn't check for that,
but it's easy enough to check for the environment variable that does the
same thing at runtime.

Prior to this commit, the code gave up immediately if it tried
unsuccessfully to use a watchdog thread.  But the alarm method is still
available.  Drop down to try it.

The next commit changes things, so that some people may find that after
that, the current name would be misleading or confusing.

The fallback method of setting a watchdog timer is to use alarm().
Prior to this commit, the code presumed that if it wasn't one of the
other possibilities, it must be this one.  This is a valid assumption
currently, but future commits will change that.  Prepare for them with
this commit.

Only one type of watchdog timer supposedly can exist.  But I can see the
possibility that a temporary glitch in the system caused a fallback type
to be used.  This commit simply makes sure that when cancelling,
anything out there gets cancelled, instead of assuming there's just one
possibility.

This enhances the code to work on cases like:

watchdog(5);
sleep 4;
watchdog(10);
sleep 8;

Prior to this commit, you had to explicitly cancel an existing watchdog
before setting a new one.

As a result of this commit, all possible wathdog variables get undef'd
early.  This makes later undef calls redundant, so they are removed.

Suggested by Tony Cook

This commit removes the existing END blocks that vary depending on the
type of timer being used, and replaces them with a single END block that
always exists, and simply calls watchdog(0).  This already cancels
whatever watchdogs are in place, doing nothing if none are.

Spotted by Tony Cook

Even though a test file should clear every timer before exit, this makes
sure it happens.