v0.5.0b1

Various speed optimizations.

• Use divide process in process_chunk()
• Add more GIL release scope
• Use split channel (Fortran memory order) I/O when appropriate (Avoid interleaving, de-interleaving)
• Fix NEON detection on ARM CPUs
• Speed improvement (compared from v0.4.0)
  • ~100% faster LQ, MQ, HQ resample on ARM processors
  • ~100% faster VHQ resample on x86_64 processors

Full Changelog: v0.5.0a1...v0.5.0b1