package:bench_press v0.4.2

• Added opt-in practical-significance and Schuirmann TOST equivalence gating
  (--threshold <delta> on bench_press run, report, and diff, plus
  defaults.threshold in bench_press.yaml, accepting relative fractions in
  [0.0, 1.0) such as 0.02 for ±2%). When threshold > 0, directional claims
  require the 95% Fieller CI to clear 1.0 + threshold or
  1.0 / (1.0 + threshold), ✅ Equivalent requires the 90% Fieller CI to lie
  completely inside [1.0 / (1.0 + threshold), 1.0 + threshold], and intervals
  straddling the equivalence boundary render as ➖ ⚪ Inconclusive (with a
  separate ⚪ N Inconclusive count in the comparison summary footer).
• Added multi-process benchmark orchestration (--processes <P> and
  defaults.processes in bench_press.yaml), compiling each
  (coordinate, runtime) cell once upfront and interleaving process launches in
  ABBA BAAB order across rounds p = 0 .. P - 1.
• Added BenchmarkMetrics.fromProcesses and ProcessVarianceComponents
  (exported from package:bench_press/bench_press.dart, serialized under
  metrics.variance_components, and surfaced as Processes and
  Between-Process Share columns in MarkdownReporter.renderSummaryTable when
  processes > 1) to decompose between-process and within-process variance
  using a drift-immune first-difference estimator (ms_within).
• Added BenchmarkConfig.defaultTrials = 16 and updated the default trials
  count across BenchmarkConfig, bench_press.yaml, and bench_press run from
  15 to 16 so that the four ABBA BAAB interleaving visits divide evenly
  into [4, 4, 4, 4] trials per visit.
• Added paired (bool, default false) to FiellerInterval and
  FiellerInterval.compute to compute covariance-adjusted Fieller confidence
  intervals (df = n - 1) when comparing equal-length paired sample vectors.
• Exported FiellerInterval and EnvironmentInfo from
  package:bench_press/bench_press.dart (alongside BenchmarkMetrics).
  External benchmark harnesses—such as browser and UI drivers that collect trial
  samples out-of-process via CDP or WebDriver rather than running an in-process
  Benchmark loop—use bench_press directly for statistical aggregation
  (BenchmarkMetrics.fromSamples), ratio confidence intervals
  (FiellerInterval.compute), and host telemetry provenance
  (EnvironmentInfo.current).
• bench_press run --target now narrows a matrix with a runtime axis to the
  requested targets instead of being silently overridden by it, and exits with a
  usage error when no coordinate matches. Coordinates without a runtime axis and
  runs that do not pass --target are unchanged.