Return raw bytes from webpages.fetch; decode HTML once in extract()

[Copilot]

webpages.fetch was returning response.text, causing requests to auto-decode bytes→str, only for downstream code (e.g. languages._from_text) to re-encode str→bytes before passing to libraries that expect bytes. This redundant round-trip risks encoding errors and wastes CPU.

Changes

• webpages.fetch() now returns response.content (bytes) instead of response.text (str)
• New webpages.html_from_bytes() decodes raw bytes → str once using trafilatura.utils.decode_file, which correctly handles charset declarations embedded in the HTML
• extract() accepts a new optional html_bytes: bytes parameter; bytes (from fetch or caller) are decoded once via html_from_bytes() before all downstream processing
• languages._from_text() no longer redundantly calls trafilatura.utils.decode_file on an already-decoded str

Usage

Callers that fetch HTML themselves can now pass raw bytes directly:

import mcmetadata
import requests

resp = requests.get(url)
results = mcmetadata.extract(url, html_bytes=resp.content)

The existing html_text: str parameter remains fully supported for backward compatibility.

• Fixes Improvement regarding content decoding/encoding #32

[rahulbot]

Rebase this off of the fix-more-unit-tests branch, which should help the unit tests pass.

[Copilot]

Rebase this off of the fix-more-unit-tests branch, which should help the unit tests pass.

Done — rebased onto fix-more-unit-tests by incorporating all its changes (commit 396fe29 and 3a7f45a): webpages.final_url() is now used in extract(), all test files use mock_fetch with requests-mock, and requests-mock was added to the test dependencies in pyproject.toml. The mock_fetch helper was updated to work with the new bytes-returning webpages.fetch() API.