fix: index html-only Gmail message bodies

[RitwijParmar]

Summary

• preserve searchable body text for HTML-only Gmail payloads by falling back from text/plain to cleaned text/html
• keep text/plain as the preferred source when multipart alternatives include both formats
• add focused regression tests for HTML-only, multipart alternative, and nested Gmail payloads
• remove an existing unused urlparse import so the touched file passes Ruff

Why

Zemail's README positions the tool as full-inbox semantic search, but many real Gmail messages are HTML-only. Before this change, _decode_body() returned an empty string for those messages, so indexing fell back to snippets rather than the actual email body. That makes retrieval/reranking weaker exactly on the long-tail emails where semantic search should help most.

The fallback is intentionally dependency-free: it strips script/style content, preserves block boundaries as newlines, unescapes entities, and normalizes whitespace before the body is embedded.

Validation

• .venv/bin/python -m pytest tests/test_gmail_client.py -q -> 3 passed
• .venv/bin/python -m ruff check zemail/gmail_client.py tests/test_gmail_client.py
• .venv/bin/python -m compileall -q zemail tests
• git diff --check