Handle malformed NXDL files without raising exceptions

[rayosborn]

Summary

Malformed or unusual NXDL files now produce logged errors from nxcheck,
nxvalidate and validate_application() instead of Python tracebacks, and a
few related validation bugs are fixed.

Robustness against malformed NXDL

• parse_nxdl(): XML syntax errors, unreadable files and a wrong root tag
  raise NeXusError.
• extends: a missing parent, a circular chain or a self-reference raises
  NeXusError, for both base classes and application definitions. A missing
  extends attribute or <group> element in an application is also handled.
• Unnamed <field>, <group>, <link> and <attribute> elements are
  skipped and reported as errors.
• xml_to_dict() skips enumeration items without a value and <dim>
  elements with a non-integer index.
• A non-integer minOccurs is ignored, as maxOccurs already was.
• lint_nxdl() reports a broken nxdl.xsd as a warning.
• GroupValidator initialises ignoreExtra* for unknown classes, and
  Validator.__repr__ tolerates a missing file path.

Data-side guards

• check_data no longer fails when @axes is longer than the signal rank
  or an axis is scalar or a group.
• check_enumeration handles array-valued fields.
• A <field> or <link> in an application definition that matches a group
  in the file is reported as an error.
• Validating from a path whose top-level group has no base class no longer
  raises AttributeError.

Other fixes

• The size of the last dimension is now checked correctly in
  check_dimensions.
• ignoreExtraFields="false" (and the other ignoreExtra* attributes) is no
  longer treated as true.
• Symbol-consistency messages are now output during application validation.
• Removed the unused Validator.get_attributes and fixed a "TThe" typo.

New tests cover each malformed-NXDL case. nxcheck and nxvalidate run
without exceptions on all files in the NeXus example data.

🤖 Generated with Claude Code