Convert name selectors and string literals to the string they denote (RFC 9535 2.3.1.2)

[gaoflow]

$["\t"] finds nothing in {"\t": "A"} while a raw tab in the selector finds it, because the parser keeps the source token of a name selector verbatim and leaves the decoding to whoever consumes it — RFC 9535, section 2.3.1.2 says the string must be converted to a member name by removing the quotes and replacing each escape sequence with the character it names. Six places downstream were each doing part of that job (normalize_json_key, the quote trimming in Value::get, prepare_regex, convert_js_path, parse_deletion_path and Pointer::key), which is also why $['\''] today returns the member named \ rather than the one named ', and why $["a"] reports its normalized path as $['"a"']. This decodes once in the parser and removes all six; \uXXXX now also accepts lower-case hex (section 2.3.1.1: the hexadecimal digits "can be either lowercase or uppercase"), surrogate pairs are joined, and normalized paths follow section 2.7 while reference/delete_by_path escape their JSON Pointers per RFC 6901.

Against the suite in rfc9535/ with filtered_cases.json emptied, failures go from 37 to 13, and the 24 that now pass are exactly the entries I removed from that file; the remaining 11 skips and 2 failures are #86, #87, #91 and the descendant-ordering case, all untouched. That is also why these survived to 1.0.6: the run recorded in results.csv as 703; 666; 2 on 2026-07-14 cannot happen in CI, since the CTS is a submodule no job checks out and rfc9535 is a binary crate that cargo nextest run never executes. I left CI alone — arming it would turn the remaining 13 red, and that is your call.

cargo test --all-features goes from 101 to 106 passing, plus cargo fmt --check, cargo clippy --workspace --all-targets --all-features -D warnings and cargo doc clean. Five existing tests asserted the old behaviour and are updated: tab_key and carr_return used documents keyed "\\t" and "\\r" (a backslash followed by a letter) where the CTS uses a real tab and CR, literal_test expected hel\'lo for 'hel\'lo', and name_sel/single_quote asserted the old un-normalized paths. I also checked the fix is not over-eager: re-adding the \\ collapse in prepare_regex breaks two CTS cases that pass on main, and escaping non-ASCII in normalized paths breaks your escaped_up_hex and surr_pairs.

Closes #88, closes #89, closes #90.

[besok]

thank you for the pr. It looks good

[besok]

looks good.