Prefixed keys in session token storage

Is your feature request related to a problem? Please describe.

I'm always frustrated when RemoveOldTokens iterates over every key in the sessions bucket and deserializes each token value to check its epoch. This becomes increasingly expensive as the number of stored sessions grows.

Describe the solution you'd like

Store two keys per token in the same bucket using prefixes:

• {pref_0}{acc} token data (for GetToken, O(1) lookup)
• {pref_1}{epoch}{acc} empty value (epoch index)

RemoveOldTokens then uses Seek({pref_1}) and iterates only over expired index entries. Both keys are written atomically in one transaction on save. A migration will be needed to rewrite existing entries into the new key format.

Describe alternatives you've considered

Keep the current implementation as-is — simpler, but suboptimal for large numbers of sessions.

Additional context

https://github.com/nspcc-dev/neofs-node/pull/3817#discussion_r2832300875

Start by tracing RemoveOldTokens and GetToken in the session token storage implementation, then inspect the sessions bucket transaction used when saving tokens. Done means token data and epoch index entries use the requested prefixes, saving writes both atomically, RemoveOldTokens scans only the epoch index, and existing entries are migrated.