v0.1.0a44: Optimize pairwise dot/norm functions

• Adds special case optimizations for certain types and/or norms to avoid computing the norm separately from the dot product.
• Provides tests to verify that these special cases work correctly.

v0.1.0a43: Allow single precision in more operations and other optimizations

• Uses single precision for masks when determining overlap in merging.
• Avoids an unneeded copy of the neurons merged.
• Allows single precision in dot_product_partially_normalized.
• Allows single precision in pair_dot_product_partially_normalized.
• Allows single precision in dot_product_normalized.
• Allows single precision in pair_dot_product_normalized.
• Allows single precision in norm.
• Allows single precision in dot_product.
• Allows single precision in dot_product_L2_normalized.
• Fixes and simplifies typing in get_neuron_dtype.

v0.1.0a42: Optimize TIFF converter

• Combines reading of data and metadata into a new tested function get_standard_tiff_data.
• Conversion proceeds using get_standard_tiff_data instead of the old get_standard_tiff_array.

v0.1.0a41: Fix Contour Bug

• Fixes a bug where contours were incorrectly constructed.
• Other minor changes.

v0.1.0a40: Some optimizations in postprocessing and other changes.

• Created a more optimized contour generating function, which should show significant speedups when extracting or merging cells.
• Various optimization for generate_hypersphere_masks, which should improve testing.
• Refactored generate_hyperdisc_masks from generate_hypersphere_masks.
• Made a minor optimization in expand_view.
• Refactor wavelet_denoising to use extract_neurons instead of keeping duplicated code.
• Fixed SGE install on Travis CI so that testing on SGE can continue.
• Redisabled testing on SGE with Travis CI due to the significant slowdown.

v0.1.0a39: Various optimizations and other minor improvements.

• Cut contour generation time by half. This should have a notable effect on postprocessing, particularly merging.
• Made some minor optimizations when merging two neurons.
• Try to avoid copying arrays when possible by checking to make sure the type is acceptable and similar.
• Various optimizations on the normalization functions like taking advantage of features after NumPy 1.7.x, etc. Should really help speed up merging sets.
• Fix up documentation of the wavelet transform and add a doctest for asymmetric scales.
• Try to allocate empty (not zeroed) arrays to speed up initialization.
• More checks to avoid computations used for debugging purposes when debugging is not enabled.
• Fix a bug where array debug recording was not occurring in the merging function.

v0.1.0a38: Minor postprocessing optimization.

• Skip computing results for array debug loggers if they don't actually serialize the result anywhere.

v0.1.0a37: Improvements to TIFF conversion.

• Switches to using tifffile for reading TIFFs and their metadata. (some things still handled by VIGRA)
• Removes pylibtiff as a dependency.
• Fixes a bug that could occur regarding descriptions from multichannel TIFFs.

v0.1.0a36: Support NumPy 1.10.x and other minor changes.

• Fixed some casting issues to support NumPy 1.10.x.
• Unpinned conda-build as the previous bugs have been fixed and included in more recent releases.
• Temporarily disable DRMAA tests on Travis CI due to problems configuring SGE on their new GCE VMs.

v0.1.0a35: Viewer improvements.

• Added intensity range selection to the color bar.
  • Selects the range that user clicks and then releases at.
  • Rescales color bar range and image to the selection.
  • Clicking on the color bar returns to the original intensity settings.
  • Selections within a existing selections are feasible.
  • All values outside the range are clipped.
  • Mouse over of points in the image still returns the correct intensity.
• Simplified other viewer code related to image selection and retrieval.