fix: encode resource identifiers as URL path segments

[bm1016bm-svg]

client.customers.find("customer#billing") currently builds a URL whose fragment is billing, so the HTTP request targets customer instead of the supplied external ID. Question marks become query strings, slashes create extra path segments, and literal percent escapes are reinterpreted. The shared URL builder affects both standard and nested resource methods.

Encode each raw path_parts element separately before joining it, leaving query serialization unchanged. Also escape standalone . and .., which quote() otherwise preserves and urljoin normalizes away. Callers supply raw identifiers, including literal % characters, rather than pre-encoded URLs.

Validation:

• The initial regression set fails against the unmodified implementation: eight URL-builder cases and four public customer lookup cases. HTTPX mock teardown also reports the four expected requests were never sent.
• Final full suite: python -m pytest -q — 463 passed (three existing deprecation warnings), Python 3.12, HTTPX 0.28.1.
• Tests assert the actual HTTPX request's raw_path and authorization header for reserved characters and dot-segment identifiers, plus nested paths and independently encoded query parameters. No live Lago API or customer data is used.
• Ruff lint and formatting checks on changed files; git diff --check.

This is separate from #410 / #421, which address missing authorization on child-resource listings.

Implemented and tested with OpenAI Codex assistance.

[cla-check-bot]

Welcome, @bm1016bm-svg!

Thanks for your first contribution!

Before we proceed with the review, please sign the Fiduciary License Agreement:

Sign the FLA

Once signed, this PR will be automatically updated.

[cla-check-bot]

Thanks, @bm1016bm-svg! 🎉

Your CLA has been signed and is now on file. We'll proceed with the review shortly.