DISubprogram incorrectly parses spFlags on LLVM 8.0+

[RyanGlScott]

#137 adapted to changes in the (rather roundabout) way DISubprogram's spFlags field is parsed, which was introduced in LLVM 8.0 (see llvm/llvm-project@adcdc1b). Unfortunately, there is a key mistake in this part of the code:

llvm-pretty-bc-parser/src/Data/LLVM/BitCode/IR/Metadata.hs

Lines 706 to 710 in f4391f7

	(diFlags0, spFlags0) <-
	if hasSPFlags then
	(,) <$> parseField r (11 + 2) numeric <*> pure 0
	else
	(,) <$> parseField r 11 numeric <*> parseField r 9 numeric

Compare this to the corresponding LLVM code:

    DISubprogram::DISPFlags SPFlags =
        HasSPFlags
            ? static_cast<DISubprogram::DISPFlags>(Record[9])
            : DISubprogram::toSPFlags(
                  /*IsLocalToUnit=*/Record[7], /*IsDefinition=*/Record[8],
                  /*IsOptimized=*/Record[14], /*Virtuality=*/Record[11]);

The llvm-pretty-bc-parser code reverses the order of cases! The LLVM code parses Record[9] if HasSPFlags is true, but in the llvm-pretty-bc-parser code, it's the other way around. This manifests itself by always having the various spFlags fields (e.g., isDefinition) always be set to 0 whenever hasSPFlags is set.