Fix #222. This PR:

• Brings the excellent work of @chriszielinski from PR Fix source positions for inlines. commonmark/cmark#298
• Fixes other issues mentioned in sourcepos broken for tables and inlines and confusing CMARK_OPT_SOURCEPOS behavior #222.

Quality assurance:

• The tests should be thourough enough.
• Also introduced new tests for extensions (strikethrough, tables).
• make test passing
• make leakcheck is passing.
• As a significant part of the added code is only run when CMARK_OPT_SOURCEPOS is set, I've tried to set and unset this flag for parser only (not for the renderer), by hardcoding this in cmark_parser_new_with_mem. In both cases, all tests pass except for a few tests failing on sourcepos mismatch (which was expected).
• We've performed test on ~250k real-world Markdown texts with the utility GFM editor and matched the detected sourcepos with the source using regexps (only for the tags we were interested in).

Performance - I wasn't able to measure any difference:

• Default setup:

  master branch:        mean = 0.8915, median = 0.8800, stdev = 0.0230
  fix-sourcepos branch: mean = 0.8875, median = 0.8800, stdev = 0.0165
  

• When CMARK_OPT_SOURCEPOS was hardcoded in cmark_parser_new_with_mem:

  master branch:        mean = 0.8935, median = 0.8900, stdev = 0.0160
  fix-sourcepos branch: mean = 0.8945, median = 0.8900, stdev = 0.0170
  

I'd be glad if this is going to be fixed, as we now have to maintain own fork with fixed cmark-gfm and a our fork of commonmarker.

And any feedback would be welcome!