Fix testnet warp sync and protocol isolation

[prop-opentensor]

Summary

• set the raw testnet chainspec protocolId to bittensor-testnet so testnet nodes stop sharing a peer set with mainnet
• add a testnet-only GRANDPA warp sync provider/verifier override for the known historical authority-set transitions that broke warp sync on testnet
• preserve the justifications needed to continue serving and consuming warp proofs on testnet

Problem

Fresh testnet warp sync is currently broken for two separate reasons.

First, the raw testnet spec still uses the mainnet protocolId, so testnet nodes can connect to the wrong peer set and receive cross-chain traffic.

Second, testnet has historical GRANDPA authority-set transitions that cannot be handled correctly by the current generic testnet warp configuration alone. In particular, the canonical pre-fragment states around blocks 4589660, 4589686, and 5534451 need explicit handling, and the 5534451 transition must be verified against pre-change set_id = 3, not 2.

Without those testnet-specific corrections, a fresh node can accept the wrong peer set, reject the relevant warp proof fragments, or stall partway through warp sync.

There is also a generic serving-side issue in the SDK for delayed GRANDPA changes. That is tracked separately in the companion SDK PR: opentensor/polkadot-sdk#22

Fix

This PR keeps the change surface testnet-specific.

• raw_spec_testfinney.json gets a distinct protocol ID so testnet discovery and gossip are isolated from mainnet.
• service.rs adds a testnet-only warp provider keyed off protocolId == bittensor-testnet.
• That provider prefixes the canonical historical transition points before any poisoned or incomplete local authority_set_changes() history.
• The verifier side uses explicit authority-set overrides for those exact transition blocks instead of relying on a global testnet set-id offset.
• Testnet no longer skips the justifications that later warp fragments depend on.

For all later normal scheduled GRANDPA changes, the provider still falls back to the backend’s live authority-set history.

Why This Is Safe

• mainnet behavior is unchanged; the existing live-chain path remains intact
• the custom warp provider only activates for protocolId = bittensor-testnet
• the hard-coded overrides are limited to three known historical testnet transition blocks
• later normal authority changes still use the normal GRANDPA scheduled-change path
• no runtime APIs or external interfaces change beyond the testnet protocolId string

Companion SDK Change

This Subtensor PR fixes the testnet-specific consumer/history side.

Official peers also need the generic SDK serving-side fix from opentensor/polkadot-sdk#22 so they can encode delayed GRANDPA change fragments correctly during warp sync.

Because that SDK change extends the warp-proof encoding, production deployment also needs Subtensor to pick up that SDK revision when it is merged. The two PRs are complementary:

• this PR corrects testnet’s historical warp verification state and peer isolation
• the SDK PR fixes delayed-change warp proof generation/verification generically for serving nodes

Notes

This patch is narrow by design. It should continue to work for future normal authority-set changes. Another code change should only be needed if testnet introduces another abnormal/manual historical GRANDPA transition that cannot be reconstructed from the normal aux history.