Unicode character at the start of a paragraph was being skipped. Add a
few simple test cases.

Fix parser not replacing NULs at the beginning.