Compiler and decompiler bug fixes

[matelgaard]

Contains two bug fixes:

• Fixes Error Saving Certain BMD Files #64, a bug with the encoding of loops in the relocation tables of BMD files, caused by the number of loops being greater than the amount that can be encoded in a single byte (FF).
  A good example of this is P5R's EN.CPK\EVENT_DATA\MESSAGE\E000\E090_001.BMD. This contains the dialogue each persona says after you fuse them ("I am Jack Frost", etc.). The number of speakers in this script is in the hundreds, since there is a speaker for each persona, and the relation table encodes the addresses for the speaker table incorrectly in this case.
  The compiler used to output the following (example still from EN.CPK\EVENT_DATA\MESSAGE\E000\E090_001.BMD from P5R):
  
  where the correct encoding is:
  
  This has been fixed to produce the correct encoding.
• Fixes the decompiler writing the minimum integer as -2147483648, and instead writes it in hexadecimal instead. This seems to have been intended, but failed previously due to a >= 16 check, which fails since the value is actually negative. This change also makes sure the decompiled script can later be recompiled without issues, since the compiler otherwise fails to parse the literal -2147483648. Perhaps this should also be fixed in its own right at some point
  There are examples from P5R's BF files where this literal appears. Here is a snippet from the decompiled EN.CPK\FIELD\HIT\FHIT_009_002.flow:
  
  From the context alone it makes sense that these would be written as 0x80000000 instead