Fix heredoc bodies losing trailing blank lines/spaces to a greedy rstrip

[agu2347]

Problem

Heredoc bodies (with preserve_heredocs=False, the flattening path) are right-stripped of all trailing newlines, tabs and spaces, so distinct sources collapse to the same value:

>>> from hcl2.utils import SerializationOptions
>>> opts = SerializationOptions(preserve_heredocs=False)
>>> hcl2.loads('a = <<MARKER\nx\n\nMARKER\n', serialization_options=opts)   # trailing blank line
{'a': '"x"'}
>>> hcl2.loads('a = <<MARKER\nx   \nMARKER\n', serialization_options=opts)  # trailing spaces
{'a': '"x"'}

Both collapse to the same "x" as a plain one-line heredoc, silently discarding a trailing blank line or trailing spaces that are part of the body per the HCL spec.

Fixes #316.

Root cause

hcl2/rules/strings.py, HeredocTemplateRule.serialize's preserve_heredocs=False branch:

heredoc = match.group(2).rstrip(self._trim_chars)  # _trim_chars = "\n\t "

rstrip with a character set removes every trailing character in that set, not just the single newline that separates the last body line from the closing marker.

Fix

Added _strip_single_trailing_newline(), which removes exactly one trailing \n (matching the established one-line-heredoc behavior every existing golden test depends on) and nothing more. Used it in HeredocTemplateRule's flattening path.

Scoped to the plain <<MARKER heredoc only, matching every reproduction case in the issue. I initially also touched the indented <<-MARKER (HeredocTrimTemplateRule) variant, but that broke test_parse_flattens_heredocs -- its rstrip call also throws away the closing marker line's own indentation (not body content), which is unrelated to this bug and has separate, pre-existing golden-test semantics. Left that variant untouched.

Testing

Added test/unit/test_heredoc_trailing_whitespace.py: a trailing blank line, two trailing blank lines, and trailing spaces now each produce distinct, non-collapsed values; a plain one-liner and an interior blank line (both already correct) are confirmed unaffected.

Confirmed each new test fails against the pre-fix code with the exact reported collapse-to-"x" behavior, and passes after the fix. Ran the full existing suite (1391 tests): all pass, zero regressions.