Fix Discord snowflake validation, add 503 retry, and Slack rate limit backoff

[Divide-By-0]

Summary

• Discord snowflake validation: When bridging Slack thread replies to Discord, the gateway can pass a Slack message timestamp (e.g., "slack 1772220754.010239") as the message_reference.message_id when the message cache misses. Discord rejects this with NUMBER_TYPE_COERCE / "not snowflake", silently dropping the message. This adds:

  1. Upfront validation that ParentID is a valid numeric snowflake
  2. Numeric-only extraction as a fallback (strip non-digit chars)
  3. Retry without threading if Discord still rejects the reference — so the message is delivered flat instead of dropped

• 503 retry at gateway level: Adds retry with exponential backoff (1s, 2s, 4s, up to 3 attempts) for transient 503 Service Unavailable responses, preventing silent message loss during Discord outages.

• Slack rate limit exponential backoff: Upgrades handleRateLimit to track consecutive rate limit hits per operation, applying RetryAfter × 2^consecutiveHits backoff. Capped at 5 minutes max wait and 10 consecutive hits before giving up. Prevents tight retry loops during reconnection bursts when matterbridge re-fetches channel/user info for many channels at once.

Test plan

• Compiled and deployed to production Slack↔Discord bridge (13 channels)
• Verified clean startup with all gateways connected
• Monitor logs for correct warn-level messages when snowflake fallback triggers
• Verify thread replies from Slack still appear on Discord (as flat messages when cache misses, as threaded replies when cache hits)
• Confirm 503 retry logging appears during Discord outages instead of immediate failure

🤖 Generated with Claude Code