ENH: BCF xml parsing: drop lxml, use xml.etree

[sem-geologist]

This PR drops usage of lxml and adopts built-in xml.etree.ElementTree

• it is not a first time when travis ci and appveyor fail/skips tests due to lxml geting inculded in/out from default anaconda library list.
• xml.etree is faster at parsing
• reduces installation complexity (so that bcf would be installed even with minimal installation)
• reduces dependency (as etree is built-in and distributed with python)

Progress of the PR

• getting working replacement of bcf.py
  • adopt complicated xpath definitions to significantly narrowed down etree xpath subset (fix overview image reading)
• remove lxml checks in other files (io/__init__.py),
• update relevant parts in documentations
• update installation scripts dropping lxml dependency (setup.py, .travis.yml, apveyor.yml ...)
• check tests and do some adjustment if needed to pass all of them, (original metadata structure is not changed going to be very minory changed, adding the XML attributes with '@Attribute' notation)
• find faulty xml example (I believe I encountered one once), and adopt regex code for its fixing prior-parsing, (containing numbers in the tags: e.g. <1>good day</1>) can't find such faulty bcf file. I have only bruker eds *.spx and project files with messy xml tags. Without working bug example this will wait until I or somebody will find such buggy bcf file.
• ready for review.

[sem-geologist]

CI checks fail not due to bcf re-implementation, but due to plotting test failures.

[ericpre]

This look alright. Can you please merge with hyperspy:RELEASE_next_minor, so that we can make sure that the bcf tests are also passing on appveyor? Since matplotlib 2.1 the plotting tests are failing... but this is another story.

[ericpre]

Thanks @sem-geologist!