test: close P2/P3 coverage gaps in leveling module

P2 — G4 (FF/SF propagation): 4 tests exercise FinishToFinish and
StartToFinish dependency types in propagation, including FF with lag.
An NRT combines all four dep types (FS, SS, FF, SF) in one project
to exercise every branch of compute_earliest_start().

P2 — G6 (hierarchical tasks): 3 tests verify propagation with nested
task structures — sibling references (phase.task_a -> phase.task_b),
cross-hierarchy absolute references (phase1.task_a -> phase2.task_b),
and an NRT with two phases each containing dependency chains.

P3 — G5 (milestone L004): 3 tests verify L004 diagnostic emission
when milestones are delayed by leveling (standard leveler, hybrid
leveler) and absence of L004 when milestones are not delayed.

P3 — G7 (calendar holidays): 3 tests verify leveling correctness with
holidays — shifted tasks land on working days, dependency propagation
respects holidays, and an NRT with multiple scattered holidays.

Leveling module now has 60 tests (up from 22 pre-fix).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: 2 people