fix(tip403): cache policy id counter

[mattsse]

Summary

• Seed the TIP-403 policyIdCounter into PolicyCache during node startup.
• Store the cached counter as Option<u64> so an unseeded cache remains explicitly unknown.
• On policyIdCounter() cache miss, fetch the authoritative value from L1 at last_l1_block, cache it, and return it.
• Keep policy record observations separate from counter updates so cached policy IDs do not become a lower-bound counter substitute.
• Prevent delayed seed/RPC results from regressing a newer cached counter.
• Serve subsequent PolicyCheck::policy_id_counter() calls from local cache state.

Root Cause

The previous implementation derived policyIdCounter() from max(cached policy id) + 1, but the policy cache can be missing policies that exist on L1. That makes the derived value only a lower bound. Keeping the authoritative value in the cache fixes the undercount without putting an L1 RPC call on every zone precompile invocation.

Impact

Contracts calling the zone TIP-403 proxy get a locally served counter after startup seeding or the first cache-miss fallback. If the counter is genuinely unknown, the proxy fetches it once from L1 and stores it for future calls instead of fabricating 2 or performing network IO every time.