kml heap-use-after-free found by fuzzer

[Kurt Schwehr]

==475427==ERROR: AddressSanitizer: heap-use-after-free on address 0x606000092900 at pc 0x0000007c611a bp 0x7ffcbfe18a90 sp 0x7ffcbfe18a88
READ of size 8 at 0x606000092900 thread T0
    #0 0x7c6119 in KMLNode::~KMLNode() third_party/gdal/ogr/ogrsf_frmts/kml/kmlnode.cpp:132:20
    #1 0x7d365d in KML::parse() third_party/gdal/ogr/ogrsf_frmts/kml/kml.cpp:134:17
    #2 0x5ec22c in OGRKMLDataSource::Open(char const*, int) third_party/gdal/ogr/ogrsf_frmts/kml/ogrkmldatasource.cpp:149:22
    #3 0x50ac05 in LLVMFuzzerTestOneInput third_party/gdal/autotest2/cpp/ogr/ogrsf_frmts/kml/kmldataset_fuzzer.cc:38:31

Minimized crash case for the fuzzer

<i/>

[Kurt Schwehr]

minimal crash case

[Even Rouault]

In 40620:

KML: make KML parser more robust (but issue cannot happen with existing external checks) (fixes #7127)

[Even Rouault]

Kurt, this is far from being a valid KML file that triggers the KML driver with the nominal code. I couldn't exactly reproduce this issue with Valgrind, but got something similar by disabling both Identify check and !poKMLFile_->isValid() test in OGRKMLDataSource::Open(). I'm not sure though of the value of removing existing checks that prevent such issues in practice. We're just giving ourselves more work than needed.

[Kurt Schwehr]

This is the kind of stuff that makes me hopeful long term sustainability with GDAL for me...