Enable compilation and caching of GRF file tables

[rdw-software]

Before this change - no caching, fully decode the TOC every time:

• About 2.1 seconds to load pay_dun00 (best case)
• About 3.6 seconds to load schg_dun01 (worst case)
• About 2.5 seconds to load prontera (average case)

After this change - on the first load (cache miss):

• About 2.5 seconds to load pay_dun00 (best case)
• About 3.9 seconds to load schg_dun01 (worst case)
• About 2.8 seconds to load prontera (average case)

After this change - every load beyond the first is a cache hit:

• About 1.1 seconds to load pay_dun00 (best case)
• About 2.5 seconds to load schg_dun01 (worst case)
• About 1.4 seconds to load prontera (average case)

Limitations:

• Cache hit/miss detection is a simple file system lookup (using the GRF file name), so name clashes may be possible
• The eviction policy is based on the cache entry's mtime, which is problematic (but probably fine, for now)
• Probably doesn't work with Arcturus/iRO alpha PAKs or similar archives, which however aren't supported at this time
• The binary format is designed so that it's 100% backwards-compatible, which wastes memory and CPU time

---

Resolves #153.