fix: swap httpx transport for requests to avoid Reddit 403

[kragent66-glitch]

Summary

Reddit's bot detection fingerprints the httpx TLS/HTTP stack and returns a 403 challenge page even with a valid reddit_session cookie. Replacing the httpx.Client with a requests.Session fixes it — urllib3/requests passes Reddit's fingerprint checks.

Closes #13 (same repro: valid cookie works with curl/urllib, httpx gets 403).

Changes

• rdt_cli/transports.py:
  • httpx.Client → requests.Session (headers + cookies copied, max_redirects=10)
  • Manual BASE_URL joining for relative URLs (requests has no base_url)
  • Explicit timeout=self.config.timeout per request (requests has no default)
  • httpx.TimeoutException/NetworkError → requests.Timeout/ConnectionError
• pyproject.toml: httpx>=0.27 → requests>=2.31 (httpx no longer imported anywhere)

Why requests instead of patching httpx?

• The 403 is driven by TLS/HTTP-stack fingerprinting, not headers — tweaking httpx headers alone doesn't reliably pass (see Valid reddit_session works with curl/urllib but httpx transport returns Reddit 403 #13's own curl-vs-httpx comparison).
• requests is a lighter, ubiquitous dependency already used transitively by most Python stacks.
• This is also what makes rdt-cli work on Termux/Android, where the httpx+brotli stack misbehaves.

Verification

• ✅ pytest tests/ → 123 passed (66 deselected)
• ✅ Live against Reddit with a real session:
  • GET /r/python/top.json?limit=2 → returns posts (no 403)
  • GET /api/v1/me → no 403, no exception