Recognize pointers wrapped in TNamed in memOutOfBounds and fix bit-to-byte comparison bug

[karoliineh]

Looking through Goblint logs for the SV-COMP MemSafety tasks revealed several issues that are fixed in this PR:

1. Pointers wrapped in TNamed were not recognized and considered as pointers
2. Bytes were compared to bits
3. In some edge cases, equality between pointers wrapped in TNamed was not working

Additional fixes:

• The comparison ID.lt casted_es casted_offs requires subtracting 1 from casted_es, but the corresponding log message about the size of casted_es was misleading because of the subtraction. The log message now displays the actual size before the subtraction to prevent confusion.

[michael-schwarz]

Nice catch, I guess this tended to kill us almost all of the time.

[sim642]

There's also a documentation problem (by me) that might've contributed to this:

analyzer/src/cdomain/value/cdomains/offset_intf.ml

Lines 92 to 96 in 153ce28

	val to_index: ?typ:GoblintCil.typ -> t -> idx
	(** Physical memory offset in bytes of the entire offset.
	Used for {!semantic_equal}.
	@param typ base type. *)

Contrary to the description, it returns the offset in bits (matching what CIL has). It was fine for semantic_equal because both would be in bits, but not for other uses (this is the only other one I think).

I'll have to open a follow-up PR to fix that. Changing the documentation would be easy, but it's not entirely the right thing: ptrdiff_t is large enough to fit offsets in bytes, but not necessarily in bits (in practice the difference might be irrelevant unless a program actually allocates more than 8th of the address space). Having to_index return bytes would be consistent with ptrdiff_ikind.