Cloud sync: health monitoring, probe interval, UI polish

[jaxernst]

Summary

Synced from lasso-cloud commits b72a0c7..bec376d.

Core improvements:

• HttpStrategy health write semantics — writes http_status separately from top-level status to prevent oscillation
• BlockSync worker WS-aware polling — reduces HTTP probe frequency when WebSocket is active (RU burn reduction)
• ProbeCoordinator 10s minimum probe interval floor
• Provider name stored on catalog refs for efficient lookup
• Profile not found returns 404 (was incorrectly 503)

Dashboard/UI:

• Distribution card shows provider names instead of raw instance IDs
• Optimistic lag smooths block height display (prevents visual sawtooth between polls)
• Effective lag prefers optimistic lag when available, uses negative block_lag convention
• Flash component redesigned as compact bottom-right toast
• Region selector uses formatted region names
• Network legend layout polish

Fixes:

• Delete dead round_robin.ex (referenced deleted ProviderPool, caused compile failure)
• Fix round-robin strategy mapping to :load_balanced in RPC socket

Skipped (diverged)

• dashboard.ex — 55 proprietary markers, fundamentally different event handling architecture
• All auth/billing/entitlements files — expected proprietary divergence

Verification

• Compiles with --warnings-as-errors
• No proprietary code leaked (grep check passed)
• mix format clean

🤖 Generated with Claude Code