6.0: remove the legacy HTML parser (edgar.files) and ChunkedDocument

[dgunning]

Why

edgar.documents is the HTML parsing engine; the legacy edgar.files package it replaced (~904K of code across 11 modules, imported by 16 modules, covered by 21 test files) is still shipped. Carrying two parsers means double maintenance and confusing failure modes — some of the remaining edgar.files code paths are the source of open bugs (#870).

Scope

Sequenced, with safety gates:

1. Finish migrating the remaining internal usages from edgar.files to edgar.documents.
2. Rebuild the new-vs-legacy parity benchmark — this gates the removal; we don't delete until the new parser demonstrably covers the old one's outputs.
3. Ship DeprecationWarnings on edgar.files entry points in a late 5.x release.
4. Delete edgar.files and ChunkedDocument in 6.0.

Removing the legacy path resolves #870 (we will verify edgar.documents handles that filer-agent HTML before closing it as fixed-by-removal).

Acceptance

• Parity benchmark green before deletion.
• Deprecation warnings shipped at least one minor release before removal.
• Upgrade-guide entry: every removed public symbol mapped to its edgar.documents replacement (Filing.document(), extractors), with a mechanical rewrite rule where one exists.