Fix null dereference comparing CompressedVector prototypes

[glitchfox]

Fixes the null dereference from GHSA-f7fw-gw7v-2h7g.

prototype_ and codecs_ are set after construction via setPrototype() /
setCodecs(), so both can legitimately be null. Every other use in
CompressedVectorNodeImpl.cpp checks for that (lines 52, 97, 178, 184, 227, 231,
244, 253) — isTypeEquivalent() was the one place that didn't, so passing a null
prototype into StructureNodeImpl::isTypeEquivalent() dereferenced it at
ni->type().

A missing prototype or codecs is now treated as "not equivalent", which is the
same answer a mismatch already gives.

codecs_ is guarded as well: it has the identical gap one line later. Our
reproducer happens to trip the prototype path first, but a later fuzzing run did
produce an input that reaches the codecs_ line, so both are needed.

Verified locally against 21c2943:

• 6 crashing inputs from the fuzzing run are all cleanly rejected after the patch
  (with E57_ERROR_HOMOGENEOUS_VIOLATION, i.e. the real structural problem is now
  reported), and all 6 still crash without it.
• No behaviour change on valid files: 46 files — the 15 in libE57Format-test-data
  plus 31 generated through the Writer API across node types, bit widths,
  cartesian/spherical geometry, optional-field combinations, scan counts and
  Image2D sections — give identical open/reject results before and after.

Happy to adjust anything. Let me know when you've added your test case and I'll
rebase if needed.

[asmaloney]

Thanks Benjamin!