v0.3.0

This release adds support for an experimented chunked HFST format that guarantees 2- and 4-byte alignment to speed up reads on the ARM platforms, and stopping crashes due to unaligned reads on 32-bit ARM platforms.

• BREAKING Rename dylib output from libhfstospell to libdivvunspell
• Introduce command line tool, divvunspell, for testing
• Implement caps heuristics for handling capitalisation in words
• Optimise tree nodes to use slightly less memory (~5% speedup in some cases)
• Add chfst support, including FFI implementation (this is not stable and will change in future)
• Refactor SpellerWorker into its own module