[Lay-quq]

Fixes #1175

With collapse_simple_statement = "Input", a call whose anonymous function argument is written across multiple lines, e.g.

pcall(function()
	return aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa == bbb
end, 5)

was expanded to pcall(\n\tfunction()\n\t\t...\n\tend,\n\t5\n) once the single-line width of the whole call (the call prefix, the collapsed function and the remaining arguments) exceeded column_width. The root cause comes from #1112: function_args_multiline_heuristic decides whether a function argument is expanded by calling should_collapse_function_body on the argument after it has been reformatted, and the reformatted node no longer has the input positions that Input mode uses to tell whether the function was single-line, so a multiline function was treated as collapsed.

This change passes the original argument's function body to that check instead (the reformatted one is still used for the width calculation). I went with this rather than checking the reformatted function for newline trivia, as done for tables in the branch below, because it keeps the exact predicate that format_function_body uses for the same node, so Always, Never and FunctionOnly are unaffected by construction and no function signatures need to change.

Tests added in tests/tests.rs, sharing one input:

• Input mode: the input from the issue stays hugged; the same function written on a single line stays single-line and the call is still expanded when too wide; a multiline function as the last argument stays hugged; a short single-line function argument stays single-line.
• Always mode: the same input, which formats identically before and after this change.

The Input test fails without the fix and passes with it. I also ran cargo test for the 9 feature combinations from ci.yml, cargo clippy --all-features -- -D warnings and cargo fmt -- --check before and after the change (rustc 1.99.1, Linux x86_64); the only difference in test results is the two new tests. Formatting the issue input and a few call variants (method calls, local ok = pcall(...), nested calls, functions in tables) gives identical output before and after in Always, Never and FunctionOnly modes, and Input output passes --verify and is stable when formatted again. I did not run the wasm, Docker or large-scale comparison workflows.