Fix Remove operation for flow mappings and sequences

[mostafa]

Problem

The Remove operation used naive line-based removal that broke YAML structure for flow styles:

• {a: valueA, b: valueB} => removing b would delete the entire line, corrupting the document
• [item1, item2, item3] => removing item2 would delete the entire line

Solution

1. Fixed yamlpath panic: Added missing node type handlers (block_mapping_pair, flow_pair, block_sequence_item) to Feature::kind()
2. Better removal strategy: Analyzes parent container to determine appropriate removal method
3. Flow style support: Extracts flow content, parses as YAML, removes target element, serializes back to flow style
4. Preserves existing behavior: Block mappings/sequences still use line-based removal where appropriate

Testing

Added a few tests covering all YAML styles and some corner cases.

Known Limitations

Added two skipped tests to show how brittle nested structure detection is. This is a TODO for future.

xref #876

[woodruffw]

I'm concerned these kinds of checks aren't sound/won't generalize, e.g. here's a parent context that will pass the first branch here but is actually meant for the second:

{[
  foo: bar,
  baz: abc,
]}

(and same for flow mappings inside of flow arrays, e.g. [{foo: bar}])

Do you have any thoughts on how to generalize these? I don't have a great sense for what would work well here -- I think testing for individual tokens without knowing where they are in the syntax tree will unfortunately be too brittle 😞

[mostafa]

I'll have a look tonight.

[mostafa]

I tried to implement this using tree-sitter, but gave up because of complexity. Instead I added two skipped tests to do it in the future (in a separate PR).

[woodruffw]

Okay, thanks! I'll have another look tomorrow.

[woodruffw]

(Just noting that I haven't forgotten about this, I've just been thinking about it -- the current yamlpatch ops are at the boundary of how much complexity I can hold in my head at once, so I've been thinking about larger architectural changes to yamlpath/new approaches entirely that would simplify our patching. Most of that is probably not worth the effort in the short-medium term, though.)

[mostafa]

Thanks for the update. It is really mind-boggling. I wish we could completely move to tree-sitter for yamlpatch, but maybe in the next version.