read_metadata lacks standard error handling

[huddlej]

Current Behavior

The read_metadata function in augur.utils allows users to pass invalid filenames or existing files with missing columns and returns an empty metadata set in each of these cases.

Expected behavior

read_metadata should raise appropriate errors when it receives invalid inputs. These invalid inputs most likely represent issues with the input data or the way the user is calling the function. By silently accepting these errors, read_metadata masks real issues with the data.

How to reproduce

See unit tests for missing and invalid files names and files with missing strain/name fields.

Possible solution

Empty, invalid, or missing filenames should raise a FileNotFound exception and not return an empty result set. This exception is raised by pandas when read_csv is called.

An existing file that doesn't have a strain or name column should raise a KeyError exception instead of an empty set. This exception corresponds to the case when a user would try to access a key of a dictionary that did not exist. Since strain or name are required for downstream analyses, we need to raise an exception when they are missing.

Additional context

This issue was raised by contribution of unit tests for the unexpected behavior in #564.