Too much blankspace gets stripped about reverse_markdown HOT 1 OPEN

I was curious as to why this didn't happen with <strong> tags. The following test does succeed:

it 'keeps whitespace surrounding bolds' do
  result = ReverseMarkdown.convert("a\n<strong>potato</strong>\nis good\nbut blankspace is better")
  expect(result).to eq "a **potato** is good but blankspace is better\n\n"
end

After some puts debuggering, I see that this one also gets its blankspace trimmed in the first pass, but then cleaner.tidy adds it back again in clean_tag_borders:

converters.convert: "\n\na**potato**is good but blankspace is better\n\n"
cleaner.tidy: "a **potato** is good but blankspace is better\n\n"

So it seems that we're losing information during convert, and assuming that ** would always be preceeded by a blankspace, which gets added in tidy (clean_tag_borders). The same technique wouldn't work for links, since the surrounding blankspace doesn't need to be there for markdown links.

My previous attempt of changing remove_border_lines to use squeeze doesn't work because it would break this test:

it 'squeezes whitespace when using paragraph tags' do
  result = ReverseMarkdown.convert("first paragraph\n<p>second paragraph</p>\nthird\nin separate lines")
  expect(result).to eq "first paragraph\n\nsecond paragraph\n\nthird in separate lines\n\n"
end

with the following diff:

 first paragraph

 second paragraph

-third in separate lines
+ third in separate lines

from reverse_markdown.