has been started without webdav server

Fix SF bug #516768 reported by Dave Wallace.

Replace use of PyMem_DEL() with PyObject_Del() on object dealloc
functions.  The use of PyMem_DEL() is incorrect for object
deallocation, because it only ever calls the low-level free().  If a
custom allocator like pymalloc is used, it needs to be called to free
the memory.

      - Collector #278: DocumentWithImages could not handle URLs with
        underscores

for dedicated webdav clients by setting the env. var
WEBDAV_SOURCE_PORT_CLIENTS

- unified output for inner links

- minor code cleanup (striping of unused spaces)

the Julian day and the week in the current year based on the Julian
calendar.

        files during installation. Fixed.

wrapped objects (caching wrapped objects being a bad thing)

        support for reading gzipped detailed logfiles

in all cases; provide a version that is not magical if we are not in Zope.
Tested with Python 2.1.1 and 2.2.1 (CVS).

        unicode support for the GlobbingLexicon

module.  My *hope* is that this takes care of the single deprecation warning
produced when running the regression tests on the trunk with Python 2.2.

Shane, I'm going to beg you to help me get my test setup cleaned up
tomorrow!

original checkin message:

    DateTime.__init__() and DateTime.__str__():  When a fractional
    number of seconds is passed in or derived, round it to the
    nearest millisecond.  Also changed str() to produce no more than
    three digits after the second's decimal point (if any).

    This ensures that DateTime->string->DateTime loses no information.
    Before this change, platform rounding errors in float<->string
    conversion could lose information in both conversion steps, so a
    DateTime object converted to string and back again didn't always
    reproduce the input object's value.  That's bad on its own, and
    testDateTime.py's testConstructor3() failed about 1% of the time
    on Linux as a result.  It doesn't fail after this patch.

    CAUTION:  Note that DateTime objects have at finest millisecond
    resolution now, even if the platform time.time() happens to have
    resolution finer than that.

Note that he also had to fiddle with the function that returned system
time, so he's got a wrapper around time.time() that rounds to 3 places
the float returned by system time.   All the testing I've done
indicates that this intermittent fixes the comparison failures.