feat(tests): add Leios endorser block log test

[mkoura]

Add test_leios_blocks.py with a test that checks the node reports the expected
Leios endorser block activity in its logs. The traces live in two namespaces,
Consensus.LeiosKernel.* and Consensus.LeiosPeer.*, and the test requires them
in two tiers. Every pool must report the messages that all block producers take
part in - EB announcements from peers and accepted announcements, acquired EB
bodies, txs and votes, its own vote and the assembled certificate. Forging,
announcing, storing and certifying an EB is reported only by the pool that won
the EB election, so those are required in the log of at least one pool. The
namespaces are anchored with \] so that Certified does not also match
CertifiedAndAnnounced.

The test is skipped outside the Dijkstra era, where the traces don't exist, and
when no Tx load generator is enabled. An EB is forged from exactly the mempool
backlog that doesn't fit into a Praos block, so with no load there is no EB, no
announcement, no votes and no certificate. For the same reason the block
production rate is checked once a cluster instance is assigned, which cannot be
a skipif marker: leios_fast produces a block every 20 seconds and gets EBs,
while the 10x faster block rate of local_fast drains the mempool.

The search waits for new blocks in rounds. Each round searches only the part of
a log that was appended since the previous round, and only for the messages that
are still missing there, so waiting for many blocks doesn't mean reading the
whole trace heavy log over and over, and a pool whose messages are all found is
not searched at all. The new search position is recorded before the search, so
that lines appended while it runs are not skipped in the next round. A stalled
chain and a log file that could not be searched are reported together with the
messages that were missing, so a failure is not just "no EB messages found".

The test cannot be marked testnets - it needs the node logs of a local cluster
and waits for many blocks - so the Leios regression run, which selects
testnets, would not run it. Add a leios marker, mark the module with it and
select it as well in scripts/test_leios.sh and in the Leios workflow.

ENABLE_TX_CENTRIFUGE and ENABLE_TX_FIREHOSE are added to the configuration,
so far they were read only by runner/regression.sh.

The expected messages were verified against the cluster logs of the 2026-08-18
leios_fast regression run - all of them are present in each of the three pool
logs. The skip paths were verified against a running cluster. The test itself has
not been run against a live Dijkstra cluster with a load generator.