Bump CURRENT_NETWORK_PROTOCOL_VERSION to 170_150 for NU7

[0xpierre-dev]

On a single-node regtest started from testnet-single-node-deploy/dockerfile
(tested at zsa-integration-demo @ 2b036fd; the same code is present on
zsa1), zebrad aborts about ten seconds after the block at the NU7
activation height (height 1 with the shipped config) is accepted:

thread 'tokio-rt-worker' panicked at zebra-network/src/protocol/external/types.rs:41:9:
Zebra does not implement the minimum specified Nu7 protocol version for Regtest { activation_heights: {Height(0): Genesis, Height(1): Nu7}, ... }

Cause: zebra-network/src/constants.rs ships
CURRENT_NETWORK_PROTOCOL_VERSION = Version(170_140), while
min_specified_for_upgrade requires 170_150 for (Testnet(regtest), Nu7)
(types.rs line 119). Once the tip reaches activation, the peer set's
chain-tip watcher calls Version::min_remote_for_height and the assert
aborts zebrad. With the shipped [state] ephemeral = true, the restart
wipes the chain back to genesis, so a client that saw its block accepted
then fails in confusing ways (template height regresses; a reissuance forks
a fresh chain and fails contextual validation, e.g.
MissingReferenceNoteOnFirstIssuance).

Repro: build and run the regtest image per the tx_tool README, mine one
block via getblocktemplate/submitblock, watch the logs for ~10 seconds.
The bundled test scenarios can complete inside that window, which may be
why CI does not catch it.

Fix: this PR bumps the constant to the NU7 Testnet value already
present as a comment in the file.

Verified locally: with this change the node survives sustained
mining/submission sessions on regtest — the same client integration suite
that deterministically failed on the unpatched image (template height
regression + MissingReferenceNoteOnFirstIssuance) passes 5/5 on the
patched one, across issuance, reissuance, transfer, burn and finalization
flows.