Bootstrap resampling is unseeded, making campaign re-runs non-reproducible

[anurag-r20]

Problem

src/bootstrap.py's resampling calls (np.random.choice, np.random.randint)
draw from the global numpy random state without ever seeding it. No caller in
examples/IBM_QAOA sets a seed either.

The practical effect, observed while building the PSS window-sticker
campaign in #84: re-running the same campaign end-to-end can yield visibly
different Pareto-frontier plots between runs — not because of a code change,
but because bootstrap resampling produces a different sample each time,
which shifts which method wins narrow near-tie regions of the frontier. It
also means the number of surviving actionable-fit rows in
run_stochastic_benchmark_pss can vary between otherwise-identical runs
(related to, but distinct from, the exact-resource-matching issue in #85).

Ask

Add an optional seed parameter (or a documented way to fix np.random's
state) through the bootstrap path, so a campaign run can be made
reproducible when that's wanted, while leaving the default behavior
unseeded for existing callers that rely on fresh randomness across runs.

Filed per review feedback on #84.