Implement LOAD UNION and LOAD FRAGMENT

[Rangi42]

Fix #632

[Rangi42]

I was going to just implement LOAD UNION, but since SECTION FRAGMENT is allowed for RAM sections, it should probably work for LOAD blocks too.

LOAD UNION is easier to implement; LOAD FRAGMENT required the extra math to adjust loadOffset and curOffset by sect->size. (This is because the offset into the LOAD section needs to take previous FRAGMENTs' size so far into account, but the offset into the parent ROM section needs to ignore them and just reserve space for the current LOAD fragment.)

Also, I just do curOffset -= loadOffset; (A) to parallel the curOffset += loadOffset; later, and (B) because manually attempting to optimize that is not an improvement: in Compiler Explorer, even at -O2, the same code is generated either way.

[Rangi42]

I'm sure this could use some more test cases and inspection of the logic. Any suggested test cases, e.g. using ALIGN, would be welcome.

[ISSOtm]

I couldn't come up with additional test cases, so I'll merge this as-is. These are mostly regression tests, anyway, so best effort.