[Added, Improved, Fixed] Joy disposition guardrails

Added

Smarter disposition guardrails

Joy now double-checks a candidate’s intent before taking final actions like disqualifying them or flagging them as not interested.
• Ambiguous replies trigger a polite confirmation instead of an immediate disqualify.
• Candidates who object to AI get a one-time human-friendly pitch and are only opted out if they decline again.
• If a candidate requests a supported language, Joy automatically switches and continues the screen instead of forwarding or disqualifying.
These guardrails reduce false negatives, keep interested talent in the funnel, and deliver a more respectful experience for candidates and recruiters alike.

Improvements

• A new “AI opt-out” tag now appears in dashboards and nightly digests, separating true opt-outs from genuine “not interested” responses for clearer analytics.
• When Joy forwards a conversation to the recruiting team, subscribers can (optionally) receive the message by email and reply directly; their response is routed straight back to the candidate thread.
  (Feature flag: messaging.forward_client.email)
• Short "yes" or "call me now" texts are no longer marked as unanswered questions, keeping nightly digests clean.
• Language handling has been expanded so screening seamlessly continues in any language enabled for the job.

Fixes

• Resolved an issue where candidate “call me” requests were silently dropped—Joy now always places the call or sends a fallback SMS.
• Confirmed that replies to forward-to-client emails now appear in the candidate conversation instead of failing with a 404.