AbstractCore v2.20.2

Automatic mail is marked so automations never answer themselves, recommendations only name
engines your install has, and media requests can run without a text model.

Added

• Automatic mail (RFC 3834): OutgoingMessage(auto_submitted=..., automation_marker=...) sends
  Auto-Submitted (auto-generated or auto-replied) and X-AbstractFramework-Automation: <marker>; EmailContext.automation_marker stamps both on every guarded_send through that
  context (auto-replied when in_reply_to is set), and on_sent reports them with the
  Message-ID. mark_automatic, auto_submitted_value and the header constants are exported from
  abstractcore.comms.email. See Email.
• Message summaries and details carry auto_submitted (the Auto-Submitted keyword, None when
  absent) and framework_marker (the X-AbstractFramework-Automation value, "" when absent),
  also in the email tools' results.
• abstractcore.providers.capability_host.CapabilityHostProvider: a provider with no text model
  that runs media-only outputs (image, video, voice, music, transcription) through the capability
  plugins. Asking it for text raises TextGenerationUnavailable (an InvalidRequestError).
• Every host profile carries engines_installed ({mlx, mlx-gen, diffusers, acestep} ->
  installed in this Python environment; module lookups only).

Changed

• Recommendations and the fresh-install seed never name an in-process engine your install does
  not have. On Apple silicon without MLX (the light install profile) the text recommendation is
  the tier's model on LM Studio (qwen/qwen3.5-9b below 24 GiB, qwen/qwen3.8-27b from 24 GiB;
  Qwen3.8 Flash-Next has no LM Studio build, so 128 GiB Macs get the 27B), with
  basis: apple_silicon_engine_fallback and the reason in tier. MLX-Gen image and video rows
  stay unset with the reason and the install command. With MLX installed the unified-memory tiers
  are unchanged.
• A capability-route write that moves a route to another provider and names no options drops the
  stored speculation request (for example an MLX route's speculation: native_mtp);
  speculation: false and every other option are kept.