fix: derive $VIDEO_NAME from the path of URL inputs

[qwist1233-cpu]

Closes #197

What
get_file_name() now parses URL inputs ("://" in the path) with urllib.parse and uses the last non-empty path segment, ignoring the query string and fragment. A URL with no path falls back to the host (192.168.0.1_554 for rtsp://192.168.0.1:554), never the credentials. The CLI logs a warning if no name could be derived at all. All three backends go through this helper, so $VIDEO_NAME is fixed for opencv, pyav and moviepy.

Why
os.path.basename() of a whole URL keeps the query string, so $VIDEO_NAME for https://host/clip.mp4?token=1.2 was clip.mp4?token=1 and the default list-scenes output name became clip.mp4?token=1-Scenes.csv, which fails on Windows. A URL with no path gave an empty name (-Scenes.csv).

How verified

• Served tests/resources/goldeneye.mp4 over a local http.server and ran scenedetect -i "http://127.0.0.1:18765/goldeneye.mp4?token=1.2" -b opencv time -e 3s list-scenes on Windows. Before: OSError: [Errno 22] Invalid argument: 'goldeneye.mp4?token=1-Scenes.csv'. After: goldeneye-Scenes.csv is written.
• New parametrized tests in tests/test_platform.py (19 pass), ruff check / ruff format --check clean, and the rest of the suite passes except the moviepy-only tests, which are skipped here because moviepy is not installed.

I picked option (c) from the discussion on the issue: derive the name from the URL path and only warn when nothing usable is left. Happy to switch to erroring out if you prefer that.

[Breakthrough]

Need to make a few minor tweaks to wording in docs/strings, but LGTM otherwise.

[Breakthrough]

Let's remove the second sentence of this message to keep it terse.

[Breakthrough]

I think this CL introduces significant enough complexity to rework this docstr entirely. Can we have a separate sub section for URL/URI-like inputs and also document what heuristic we use to detect them?

[qwist1233-cpu]

Both done.

• The warning is now one sentence; the "set an explicit filename" hint is gone.
• get_file_name has a dedicated URL and URI-like inputs section. It names the heuristic (anything containing ://, i.e. is_url) and why URLs are split by component instead of by path separator, then covers the two cases separately: last non-empty path segment (percent-decoded), and the host fallback when there is no path. It also states the two things that were implicit before — userinfo is never used, and include_extension=False does not strip anything off a host, since the last . there is a domain separator. Arguments:/Returns: sections match the rest of the module.

One behaviour change came out of writing that: a URL with a port but no host (rtsp://:554) used to return _554. It now returns an empty string, which is what the docstring promises for "no usable name". Two cases added to test_get_file_name_without_extension for it.

pytest tests/test_platform.py passes (21 tests), ruff check and ruff format --check clean on the three changed files.