Reader writer fixes

[camFoltz]

This PR fixes a lot of the issues that @ieivanov brought up on the issue page.

Specifically:

Writer:

• Fixes how users define contrast limits. Can be a list of tuples with (start, end, min, max) or (start, end) where min/max will be inferred from the datatype
• Changes default array name to arr_0 instead of array
• Changes specifiyng [0,N] as a range for writing data and instead implements pythonic slice(0, N)
• changed default colormap to 'FFFFFF' and opened up an issue in the napari-ome-zarr plugin
• ome-metadata sets on the first channel as active (for viewing in napari) rest are by default inactive

Reader:

• automatic filetype parsing if the data_type is not explicitly provided by the user
• zarrfile reader can handle the old array name and also arr_0
• closes TiffFile after opening to avoid a warning and to properly clean up the file.

Other:

• updated one np.float in background_estimator to np.float64 in order to avoid numpy deprecation

[camFoltz]

@ieivanov do you think you can take a quick look here so we can merge? I have implemented and tested all of your suggestions (thanks!)

[lihaoyeh]

I can use this branch to load data well. When writing, I faced some issues in the specification of clims.

Here is the code block that I use to write the uPTI data.

writer = WaveorderWriter(output_path, hcs=False, hcs_meta=None, verbose=True)
writer.create_zarr_root('uPTI_physical_2.zarr')
position = 0
chan_names_phys = ['Phase3D', 'Retardance3D', 'Orientation', 'Inclination', 'Optic_sign']
phys_data_array = np.transpose(np.array([phase_PT, np.abs(retardance_pr_PT[0]), azimuth[0], theta[0], p_mat_map]),(0,3,1,2))[np.newaxis,...]
data_shape_phys = phys_data_array.shape
chunk_size_phys = (1,1,1)+phys_data_array.shape[3:]
dtype = 'float32'
writer.init_array(position, data_shape_phys, chunk_size_phys, chan_names_phys, dtype,  position_name='Stitched_physical', overwrite=True)
writer.write(phys_data_array, p=position)

I did not encounter an issue using the previous version of the writer. When running this code with this branch, I encounter the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-19-5c76b46745f2> in <module>
     11 # clims = [(-0.15,0.15),(0,0.012),(0,np.pi),(0,np.pi),(0,1)]
     12 dtype = 'float32'
---> 13 writer.init_array(position, data_shape_phys, chunk_size_phys, chan_names_phys, dtype,  position_name='Stitched_physical', overwrite=True)
     14 writer.write(phys_data_array, p=position)

~/project/Polscope/waveorder/waveorder/io/writer.py in init_array(self, position, data_shape, chunk_size, chan_names, dtype, clims, position_name, overwrite)
    145 
    146         self.sub_writer.create_position(position, pos_name)
--> 147         self.sub_writer.init_array(data_shape, chunk_size, dtype, chan_names, clims, overwrite)
    148 
    149     def write(self, data, p, t=None, c=None, z=None):

~/project/Polscope/waveorder/waveorder/io/writer_structures.py in init_array(self, data_shape, chunk_size, dtype, chan_names, clims, overwrite)
     61             raise TypeError('dtype must be instance of either np.dtype or str')
     62 
---> 63         self.set_channel_attributes(chan_names, clims)
     64         self.current_pos_group.zeros('arr_0', shape=data_shape, chunks=chunk_size, dtype=dtype,
     65                            compressor=self.__compressor, overwrite=overwrite)

~/project/Polscope/waveorder/waveorder/io/writer_structures.py in set_channel_attributes(self, chan_names, clims)
    337         for i in range(len(chan_names)):
    338 
--> 339             if len(clims[i]) == 2:
    340                 if 'float' in self.dtype.name:
    341                     clim = (clims[i][0], clims[i][1], -1000, 1000)

TypeError: 'NoneType' object is not subscriptable

It seems to mean that we need to specify clims in this version of the writer. I think that the previous version of the writer is convenient in that you don't have to specify contrast limits for individual channels (it auto-scales to min and max). Then, I supply the clims with the following code piece:
clims = [(np.min(array),np.max(array)) for array in phys_data_array[0]]
I then got another error:
TypeError: Object of type float32 is not JSON serializable
This is because the type of the number in the clims tuple is numpy.float32 and that is not serializable. I then converted those numbers to float. It passes.

With these tests, there are a few questions to be answered:

1. Do we want the writer to identify clims from images if it is not specified? If no, the current implementation is fine.
2. We can maybe add some checks on the type of numbers in clims or convert the number into float before doing further writing processes.

I do like the part of this version that uses "FFFFFF" as the default colormap. It visualizes images with a brighter contrast to start with.

[camFoltz]

Thanks for catching this @lihaoyeh. We want their to be default clims if the user doesn't specify any. It seems like there is a slight bug and I will fix it.

[camFoltz]

@lihaoyeh I also added the catch for the numpy datatypes that aren't serializable and now the clims will automatically be converted to regular floats

[lihaoyeh]

I have tested this new version and it works for my test case. I have resolved the conflicts and merged the master into this branch. Please go ahead and merge it to the master.