Add API to load trajectories or databases, one molecule at a time

[theochem]

Add API to load trajectories or databases, one molecule at a time

[tovrstra]

Issue #26 is essential the same as this one.

[tovrstra]

Another use case is Reaction Trajectory & IRC (Intrinsic Reaction Coordinate) See https://github.com/stvogt/kudi and http://gaussian.com/irc/

[FarnazH]

Covering trajectory stored in:

• xyz file
• sdf file
• Gaussian log file (IRC, geometry optimization, ...) (load only)

API in iodata/iodata.py

• load_many(filename) function returning a list of IOData.
• load_one(filename) function instead of IOData.from_file.
• dump_many(filename) function.
• dump_one(filename) function instead of IOData.to_file.

API in file format modules:

• load and dump functions take a file object and load or dump one IOData object at a time.
• Not all formats can be turned into trajectories because the end of the file is not easily recognized. We should raise an Error when using load_trajectory or dump_trajectory on an inappropriate file format.
• There should be a possibility for specialized load_trajectory functions, e.g. fchk and Gaussian log have special formats that are not simply a concatenation of geometries.

[tovrstra]

Here is a list of steps we need to take to support trajectories:

• Change IOData.from_file to load_one. Change IOData.to_file to dump_one.
• Change all load functions to take a file object as argument, instead of a file. They should also only load data until they have found enough to make one IOData object. When hitting the end of a file, it should raise a StopIteration. See https://docs.python.org/3.7/library/exceptions.html#StopIteration. This also means load_one should open the file object.
• Similar to previous but for dump_one
• Write load_many and dump_many functions.

[tovrstra]

@evohringer @fwmeng88 I'm working on this, so it does not hold you back. It will be a rather big change, but I'll stick to the essentials.