Optimize single-tile reads

[thejohnhoffer]

This resolves issue #347. The _bulkReadable function detects when a single read should match an IFD buffer byte-for-byte. Instead of extracting every single pixel individually, this PR copies the IFD buffer to a single typed array whenever possible.

For our test case loading unsigned 16-bit 1024x1024 tiles, I observe this optimization is 4x faster thanv2.0.7.

I have written the code to optimize only when possible:

• Requesting a single non-interlaced tile without color planes
• Ensuring the byte order of the device matches the image byte order

Testing

• I have not yet tested with multiple formats / byte counts.
• I have not yet tested on a big endian OS.

[potion-cellar]

+1