Extract the analysis layer, centralize defaults, add a panel view

[fpedd]

Move the exact happens-before queries out of primitives/ into a new
analysis/ package on both language sides: conflicts, linearize,
pressure, and the clock helpers in Python; antichain, closure,
linearize, placement, and clock.hpp (ex clock_rows.hpp) in C++.
primitives/ is now strictly the System/Memory/Pool/Allocation
hierarchy, and the conflict kernels leave allocators/first_fit for the
new analysis/conflicts.{hpp,cpp}, gaining budgeted faces
(get_conflicts, get_conflict_degrees) that give up (None) instead of
stalling once a work budget is exceeded.

Make the _cpp boundary default-free: every nanobind parameter is
explicit and Python owns all defaults in common/constants.py
(DEFAULT_TIMEOUT/SEED/WORK_BUDGET/CLOSURE_CAP plus the units absorbed
from common/units.py), so a value can never drift between the
languages. Timeouts follow one None-disables convention end to end —
float | None over std::optional, with make_deadline throwing on
a non-positive or non-finite set value even for raw-binding callers —
via the new common/deadline helpers mirrored in Python and C++.
OmniAllocator gains a validated linearize_budget (None = always
decide); minimalloc rounds its whole-second budget up instead of
truncating; the supermalloc deadline starts after partition setup so
the budget covers greedy + search only.

Rework plot_allocation around explicit views: the new default "panel"
draws each memory once over a happens-before-monotone virtual time
(exact for scalar or linearizable lifetimes, else a sound
clock-component-sum projection annotated with its conflict coverage),
and "lanes" keeps the per-thread projection with an optional max_lanes
top-k cap by definite-occupancy peak.