fix(consensus): guard against ln(1-f) panic when active_slot_coeff >= 1
assert_leader_stake currently computes c = ln(1 - active_slot_coeff)
unconditionally. For genesis params with activeSlotsCoeff = 1.0
(used by the antithesis short-epoch fixture in
lambdasistemi/amaru-bootstrap and by the live consumer test in
lambdasistemi/amaru-bootstrap#35) this is ln(0), and pallas-math's
FixedDecimal::ln panics with "ln of a value in (-inf,0] is undefined".
When f = 1, every slot is leader-elected with probability 1 — the
leader-stake assertion is trivially satisfied for any pool with
non-zero relative stake. Mirroring the pre-existing zero-active-stake
guard added in b69fa13e, short-circuit to Ok before the ln call.
The math is continuous at f -> 1 (exp(x*c) with c -> -inf collapses
to 0, which is < recip_q, so the ordering is LT = Ok), so the guard
does not change behaviour for any f < 1.
Surfaced by the live amaru-run consumer test in
lambdasistemi/amaru-bootstrap#35 once
lambdasistemi/amaru#2 (saturating epoch subtraction) let header
validation reach this code path on a cold-start short-epoch chain.
Signed-off-by: Paolo Veronelli <[email protected]>
Signed-off-by: paolino <[email protected]>