[backmeupplz]

Two fields Neynar returns on cast objects that the v2 API leaves empty today. A client building reply threads needs both: whether the viewer follows each author and already liked or recast each cast, and which conversation a reply belongs to. Without them it makes extra requests for every cast on the page.

viewer_fid

viewer_fid was already parsed in dispatch but never used. It now works on:

• /v2/farcaster/feed/following, /v2/farcaster/feed (following and trending), /v2/farcaster/feed/trending and its aliases for_you and topic
• /v2/farcaster/feed/channels, /v2/farcaster/feed/parent_urls
• /v2/farcaster/feed/user/casts
• /v2/farcaster/cast, /v2/farcaster/cast/bulk, /v2/farcaster/casts
• /v2/farcaster/cast/conversation (every cast in the tree)

With it, each cast gets viewer_context: {"liked": bool, "recasted": bool}. The author and each mentioned profile get viewer_context: {"following": bool, "followed_by": bool, "blocking": bool, "blocked_by": bool}. These are the CastViewerContext and ViewerContext types already in types.rs. blocking/blocked_by come from block links, the same links /v2/farcaster/block/list reads.

Each value is one point lookup (ReactionStore::get_reaction_add, LinkStore::get_link_add), behind two new HubQueryHandler methods, has_reaction and has_link. That is 2 lookups per cast and 4 per user. Without viewer_fid nothing is looked up and viewer_context is left out, as before.

thread_hash

thread_hash is always on, set in message_to_cast, so every endpoint that returns casts has it:

• a top-level cast (no parent, or a parent URL) has its own hash, with no lookup;
• a reply has the hash of the top-most ancestor, found by following parent_cast_id through get_cast_by_hash with the parent's FID. That is one direct cast store read per level, so a depth-d reply costs d reads;
• if an ancestor is missing, or the chain is longer than 64 hops (MAX_THREAD_HOPS), the field is left out, as it is today.

There is no cache. Ancestors don't change, so one could be added later if conversation endpoints show the cost.

Tests

• test_thread_hash_walks_to_the_root: a root, a reply and a reply to the reply all get the root's hash. A missing ancestor gives none. A self-parented cast stops at the hop limit.
• test_viewer_context_only_with_viewer_fid: /cast with viewer_fid returns the cast's and users' viewer_context from the viewer's likes, follows and blocks. Without viewer_fid there is no viewer_context on the cast, the author or mentioned profiles.
• test_has_link_and_has_reaction (server_tests): the real store lookups across two shards. A follow and a like are found. A reverse follow, a block and a removed recast are not.

cargo test --lib: everything passes except 4 storage tests (test_storage_slot_from_rent_event, test_storage_slot_with_mix_of_units, test_user_with_low_total_storage_cannot_lend, test_commit_username_proof_messages). Those fail the same way on current main. cargo fmt --all --check is clean, and clippy reports nothing in the new code. I haven't run it on a synced node.