fix(server): serve product follow aliases and stamp statement expiry

[replghost]

Problem

Epoca's live A/B against Paseo services (paritytech/epoca#174, evidence in the #168 rollout hold) showed the embedded runtime is not live-equivalent to the shipping legacy host loop, in two ways. Both were reproduced against the live People-next node and hermetically against the shipped wasm bundle:

1. Rejected papi follow ids → zero Browse products / chain-query timeouts. Every papi product (via @novasamatech/product-sdk) issues chainHead operations carrying the provider's synthetic follow subscription id (follow_0), never the follow subscribe request id. ChainConnection::require_remote_follow strict-matches its own keys and fails with unknown follow subscription id "cN:follow_0", so papi clients never initialize. The host-api protocol documents the field as "the follow subscription ID from the product-SDK (opaque string)", and the native host coordinators (e.g. UserAgentKit's MobileChainFollowCoordinator, Epoca's legacy actor) already resolve it to the session's active follow.
2. Expiry-less statement submissions rejected live → Statement Store GenericError. Products create proofs for statements without an expiry field and expect the host to stamp one. Deployed statement stores (People-next runs the legacy expiry dialect: u64, unix seconds in the upper 32 bits) refuse expiry-less statements — the node answers {"status":"invalid","reason":"alreadyExpired"} and smoldot's statement store phrases the same rejection as Invalid statement encoding. The runtime signed and submitted the product's statement as-is.

Change

• chain_runtime.rs: ensure_follow_context now resolves the operation's follow id through ChainConnection::resolve_follow_key. An exact key wins; an unknown session-scoped id is treated as an opaque alias and bound to the session's most recent unaliased follow. Aliases never resolve across session scopes (cN: prefixes), are dropped with their follow, and are bounded per connection. Unknown ids with no active session follow keep the existing error.
• statement_store: proof creation (create_proof and create_proof_authorized) stamps a one-hour legacy-dialect expiry (priority zero) when the product omitted it, signs the stamped fields, and retains the host-assembled SignedStatement keyed by its proof (entry-bounded bridge in RuntimeServices). Submission prefers those exact retained bytes over the product's pre-proof re-assembly, so the submitted fields always match the proof's signature material; the read-after-write cache serves the statement the store actually accepted.

Validation

• New regression tests: header_request_binds_product_follow_alias_within_session, header_request_rejects_product_follow_alias_across_sessions, statement_store_create_proof_stamps_missing_expiry_and_retains_statement, statement_store_submit_prefers_host_assembled_statement.
• cargo test -p truapi-server --lib: 400 passed. cargo clippy -p truapi-server --all-targets clean; cargo check -p truapi-server --target wasm32-unknown-unknown clean; cargo fmt applied.
• Live-node semantics pinned by direct statement_submit probes against wss://paseo-people-next-system-rpc.polkadot.io: upstream priority: u32 encoding fails decode (Could not decode Field, variant doesn't exist), expiry-less statements return alreadyExpired, future-expiry legacy statements proceed to proof verification.
• Epoca currently works around both defects by routing the affected tag families to its legacy loop (paritytech/epoca#209); that fallback is removable once this lands and UserAgentKit re-vendors (its truapi-server pin predates current main and needs its own adaptation — useragent-truapi-runtime does not compile against main today).

[pgherveou]

same as #317

reopen if got that wrong