fix(reformatter): ensure idempotent formatting for dictionary values with line continuation

[jdymitarai]

Summary

Fixes #1217 where yapf repeatedly added spaces before a line continuation backslash (\) in dictionary values on subsequent formatting runs (under styles enabling INDENT_DICTIONARY_VALUE, such as google). Also resolves an AssertionError when dictionary values consisting of a single leaf were preceded by a line continuation backslash.

Cause & Analysis

1. Continuation node parentage: In continuation_splicer.py, continuation nodes were inserted into pytree nodes via node.children.insert() directly instead of node.insert_child(), leaving continuation_node.parent as None. Any subsequent AST transformations or sibling queries (next_sibling, replace) on these nodes could fail or raise an AssertionError.
2. Continuation in dictionary values: In subtype_assigner.py, _InsertPseudoParentheses only checked grammar_token.COMMENT when looking for tokens inserted before a value leaf, missing format_token.CONTINUATION. As a result, single-leaf values preceded by a backslash failed to be encompassed into an atom node alongside the continuation marker.
3. Redundant whitespace emission: In reformatter.py, _FormatFinalLines appends a space for pseudo-parentheses if the succeeding token's whitespace_prefix does not start with whitespace. Because continuation tokens hold their leading whitespace in their token value (' \') rather than whitespace_prefix, _FormatFinalLines emitted an additional space before the continuation token on every formatting run, causing infinite drift.
4. Token space calculation: In logical_line.py, continuation tokens were checked after pseudo-parentheses in _SpaceRequiredBetween, allowing pseudo scope opening to request an unnecessary space before continuation markers.

Solution

• Use node.insert_child(...) in continuation_splicer.py to maintain consistent parent links.
• Handle format_token.CONTINUATION alongside grammar_token.COMMENT in _InsertPseudoParentheses.
• Check not tok.next_token.value.startswith(' ') in _FormatFinalLines before appending spaces for pseudo tokens.
• Prioritize continuation token spacing checks in logical_line._SpaceRequiredBetween.
• Add unit tests in reformatter_basic_test.py verifying multiple-pass idempotency and single-leaf dictionary value continuation handling.