RegexMatchSpan with sep="" concatenates words with sep="(space)"

[HiromuHota]

Describe the bug
A clear and concise description of what the bug is.

A sentence "123 456 789" is parsed and gets three words "123", "456", and "789".
I'd like to match a number like

RegexMatchSpan(rgx=r"\d{9}", sep="")

but sep="" has no effect.

To Reproduce
Steps to reproduce the behavior:

1. Have a sentence "123 456 789"
2. Parse it
3. Try to match it with RegexMatchSpan(rgx=r"\d{9}", sep="")

Expected behavior
A clear and concise description of what you expected to happen.

RegexMatchSpan(rgx=r"\d{9}", sep="") matches a sentence of "123 456 789".

Environment (please complete the following information):

• Fonduer Version: 0.6.2

Additional context
Add any other context about the problem here.

I think the root cause of this issue is the following implementation.

fonduer/src/fonduer/candidates/models/span_mention.py

Lines 140 to 159 in f866248

	def get_attrib_span(self, a, sep=" "):
	"""Get the span of sentence attribute *a*.
	Intuitively, like calling::
	sep.join(span.a)
	:param a: The attribute to get a span for.
	:type a: str
	:param sep: The separator to use for the join.
	:type sep: str
	:return: The joined tokens, or text if a="words".
	:rtype: str
	"""
	# NOTE: Special behavior for words currently (due to correspondence
	# with char_offsets)
	if a == "words":
	return self.sentence.text[self.char_start : self.char_end + 1]
	else:
	return sep.join(self.get_attrib_tokens(a))

where a is words by default.

[HiromuHota]

This "Special behavior" dates back to snorkel-team/snorkel@7b57927.