Preserve non-canonical numeric nested JSON keys

[upuddu]

HTTPie's nested-JSON syntax decides whether a key segment is an array index or a text key by calling int() on it. int() is lenient - it also accepts 007, 1_000, +5, and whitespace-padded values - so numeric-looking object keys were silently rewritten. http pie.dev/post 007=1 sent {"7": "1"} instead of {"007": "1"}, quietly dropping the leading zeros; 1_000 became 1000 and +5 became 5.

This only treats a segment as a number when it's written in canonical integer form, keeping everything else as verbatim text. Canonical values including negative indexes like foo[-10] are unaffected. Added coverage to test_nested_json_syntax and a changelog entry.

[upuddu]

The failing checks here are pre-existing on master (5b604c3) and unrelated to this PR, which only touches httpie/cli/nested_json/parse.py and its test:

• test_encoding.py::test_terminal_output_*_charset_detection[big5-…] fails because current charset_normalizer (3.4.x) detects the big5 sample as EUC-KR. It also fails on unmodified master.
• test_auth.py::test_digest_auth[https-…] fails intermittently with Context has already been used to create a Connection (urllib3/pytest-httpbin SSLContext reuse).
• The 3.7 matrix legs fail at setup because GitHub runners no longer provide Python 3.7; the macos-13 legs time out at the runner level.

All added tests pass locally (pytest tests/test_json.py → 295 passed). Repo-level fixes for the above (dependency pins / dropping 3.7 / SSLContext test fix) belong in separate infra PRs, not here. Happy to rebase once master's CI baseline is repaired.