Add regtest block relay test and remove sync workaround

[oxarbitrage]

Depends on ZcashFoundation/zebra#10403 (merged)

• util.py: Remove the node-restart workaround from rebuild_cache(). Sync now relies solely on sync_blocks(), which is the correct behavior now that the Zebra fix has landed.
• regtest_block_relay.py: New regression test for bug(network): Zebra broadcasting in regtest mode ZcashFoundation/zebra#10332 (comment). Three nodes in a linear A -- B -- C topology; node A mines 10 blocks; asserts all nodes reach height 10. This is the minimal reproduction case from the issue.

[oxarbitrage]

Zebra fix (ZcashFoundation/zebra#10403) is now merged, so this is ready for review.

[oxarbitrage]

Rebased onto current main (was 12 commits behind; last full CI predated #138).

Conflict was in util.py's initialize_chain cache builder, where main had since added the zebra-broadcast workaround (sync_blocks_with_reconnect + stop/restart all zebrads, zebra #10329/#10332). Resolved in favor of this PR's intent — removing the workaround now that ZcashFoundation/zebra#10403 has merged — so cache sync relies solely on sync_blocks(rpcs) (60s timeout: fails fast rather than hanging if the broadcast fix isn't present). Also dropped the now-orphaned sync_blocks_with_reconnect helper, which only existed to serve that workaround.

The new regtest_block_relay.py is in NEW_SCRIPTS with cache_behavior='clean', so it does not trigger the 8-node cache pre-populate that #138 just fixed. This CI run is the real check that zebra#10403 is in the test binaries.