test: grammar-conformance differential (transcribe docs/grammar.md, diff against the parser)

[enomado]

What

ron has ~450 correctness tests but nothing checks that the parser agrees with the
grammar it documents in
docs/grammar.md —
there is no grammar-conformance test.

This adds one. conformance/ is an isolated sub-crate — the same trick as
fuzz/ (its own [workspace]), so pest never enters ron's own dependency
graph or the MSRV job. It transcribes the EBNF verbatim into a PEG acceptor
(conformance/src/ron.pest) and diffs it, input by input, against the parser. A
disagreement is either a parser bug or a grammar-doc bug.

The oracle for "does ron accept this as one syntactically-valid value?" is
Deserializer::from_str -> IgnoredAny::deserialize -> end(): pure syntactic
validation, whole input consumed, no data model imposed. (Value can't be the
oracle — it lacks struct/enum/range/byte variants and would reject or mangle them.)

What it found

Building the differential surfaced seven places where docs/grammar.md
disagrees with the parser — in every case the parser is right and the doc is
wrong. They are filed as #614 and pinned here in KNOWN_DOC_BUGS:

#	input	doc	parser	gap
A	"a\c"	accept	reject	a lone \ must start a valid escape
B	'\n'	reject	accept	char takes the full string escape set
C	"�"	accept	reject	\u requires { 1..=6 hex }
D	1 ..2	accept	reject	no whitespace around range operators
E	'''	reject	accept	a bare ' is accepted as char content
F	1e_+0	reject	accept	_ allowed before the exponent sign
G	#![enable(..)]	reject	accept	leading whitespace before extensions

Each row is a self-checking assertion: the transcription is faithful to the
doc, the parser behaves as claimed, and the two disagree — a wrong expectation
fails loudly instead of passing silently. Everywhere else (the CONFORMANT
battery, spanning every production family) the transcription and the parser agree.

Inputs ron rejects for semantic reasons (911u8, -1u8, Some()) are pinned
separately in SEMANTIC_NOT_GRAMMAR, so the differential does not mistake them for
conformance gaps.

CI

A dedicated Conformance job runs cd conformance && cargo test, isolated from the
main test / MSRV matrix (pest is pulled into this job only).

Note

This is the executable evidence behind #614. If useful, I'm happy to also open the
grammar.md fix PR; once the doc is corrected the KNOWN_DOC_BUGS entries fold
into the conformant battery and the divergence set goes to zero, leaving the test
as a guard against any new drift.

[juntyr]

@enomado Thank you for all of your PRs today! I will have a look at them over the coming days once I have some free time