v0.1.1

Fixed

• SynthesizedUnitaryBlock and the other Quantum Shannon Decomposition
  paths are accurate to ~1e-13 on structured unitaries (DFT, Hadamard,
  phase-decorated targets) up to 7 qubits, where a 6-qubit DFT was far off:
  the cosine–sine decomposition divided by small sines, and the eigenvalue
  and diagonal tolerances merged distinct values.
• Synthesized unitaries are phase-exact. The global phase is kept down to
  1e-15, so a synthesized block under ControlledBlock (for example inside
  QPE) matches the controlled target exactly instead of carrying a spurious
  relative phase.
• The default thread count no longer starts an OpenMP worker on every
  hardware thread, which starved the main thread. With neither
  QARP_NUM_THREADS nor OMP_NUM_THREADS set, every layer uses the physical
  cores in the process's CPU affinity, capped one below its logical CPUs, and
  a container CPU limit (any cgroup v1 or v2 quota) lowers it further. Under
  WSL and some virtual machines the reported topology undercounts physical
  cores; set QARP_NUM_THREADS there.

Changed

• CITATION.cff names the arXiv paper as the preferred citation, with the
  Zenodo concept DOI for the software.