SoloCrawl 1.2.0

SoloCrawl 1.2.0 hardens fetching, proxy recovery and URL safety after an independent code review.

Changes

• Recover failed proxy endpoints after a 60 second cooldown instead of disabling them for the life
  of the process; a 429/503 from the target site no longer marks the proxy as failed.
• Block shared and other non-global addresses, including CGNAT/Tailscale 100.64.0.0/10.
• Give up instead of waiting when Retry-After is longer than the request timeout.
• Use the Playwright fallback only for HTML responses (not short JSON or PDF), and return the
  rendered page when the browser is forced even if its content is short.
• Load robots.txt for different hosts concurrently instead of one at a time.
• Avoid repeated DNS checks for the same URL during a fetch.
• Add select_providers(config, sources=...) in the search core; the MCP and CLI shells use it.
• Log the exception type when a browser fetch fails.
• Remove dead code (legacy duckduckgo_search import, signal-handler cleanup that never ran) and
  the stale requirements.txt.

Compatibility notes

• ProxyEndpoint no longer has a healthy field; failure state is tracked by ProxyPool.
• The legacy duckduckgo_search package is no longer used as a fallback; ddgs is required (as
  already declared in the dependencies).
• Install from pyproject.toml; requirements.txt has been removed.
• The five public MCP tool names and signatures remain unchanged.

Verification

The full verification runner passed: Ruff, formatting, Pyright, offline fixture tests (447), MCP
protocol, local Chromium, dependency consistency, vulnerability audit, wheel/sdist build and clean
base/browser/all installations. All ten package registries, the public search providers, and live
federation, fetch, research and batch passed. SearXNG was skipped because no instance was
configured.