Add an as_json flag to dumps()/dump() for guaranteed RFC 8259-compliant output #158

Filing this as the follow-up that came out of #151, per the discussion there.

dumps() has several defaults that make its output legal JSON5 but not legal JSON:

• identifier keys are emitted unquoted ({foo: 1})
• trailing commas are added when indenting
• NaN/Infinity literals are allowed (allow_nan=True by default)

The dump() docstring claim that quote_keys=True, trailing_commas=False "should produce exactly the same output as json.dump(obj, fp)" turned out to be misleading, because the json module also refuses non-finite floats where the JSON grammar has no spelling for them. #155 fixed the CLI side of this (--as-json now rejects non-finite numbers), but the underlying API still makes callers compose the guarantee themselves.

Proposal: an as_json flag on dumps()/dump() that bundles the JSON-compliance guarantees:

• quote all object keys
• no trailing commas
• reject NaN/Infinity (allow_nan=False behavior)

with the explicit goal from #151 being: the output is legal JSON per RFC 8259 and evaluates to the same values — not byte-for-byte identity with the json module's output.

The CLI's --as-json mode could then route through dumps(as_json=True) so the library and the CLI share one definition of "JSON output" (today the CLI hand-sets the three kwargs and the flag's semantics live partly in tool.py).

Happy to take a crack at this if that's useful, though it looked like a largeish change touching the docstring compatibility claims too.