ci: grant issues:write to pr-cleanup workflow

[Qazalbash]

Removing a label requires issues:write, not just pull-requests:write, which was causing the label-removal step to fail with a 403.

[github-actions]

Benchmark report

this PR pr-cleanup-update at 64bbac91 vs baseline master at 5f9090f4

- run time:     1 slower, 1 faster
  compile time: unchanged across 32 benchmarks

Significant changes (2)

                         ──────── run time ───────     ───── compile time ─────
  benchmark              baseline  this PR       Δ     baseline  this PR      Δ
───────────────────────────────────────────────────────────────────────────────
+ categorical_log_prob     2.4 ms   2.1 ms  -12.0%      70.3 ms  70.3 ms  +0.1%
- gamma_log_prob           2.1 ms   2.3 ms   +7.3%       2.22 s   2.43 s  +9.3%

_{Red is slower, green is faster; a row is coloured by the worse of its two columns. A delta in parentheses cleared the threshold on a measurement below the resolution floor, so it is shown without being called a change. † marks a benchmark that could not be compared — see below.}

Full results

distributions

                                 ──────── run time ────────     ────── compile time ──────
  benchmark                      baseline  this PR        Δ     baseline   this PR       Δ
──────────────────────────────────────────────────────────────────────────────────────────
  biject_to_constraints            4.0 ms   4.2 ms    +3.3%     379.6 ms  378.4 ms   -0.3%
+ categorical_log_prob             2.4 ms   2.1 ms   -12.0%      70.3 ms   70.3 ms   +0.1%
  dirichlet_log_prob               717 µs   710 µs    -1.0%     425.3 ms  406.2 ms   -4.5%
  dirichlet_sample                45.6 ms  45.5 ms    -0.3%     883.2 ms  904.3 ms   +2.4%
- gamma_log_prob                   2.1 ms   2.3 ms    +7.3%       2.22 s    2.43 s   +9.3%
  gamma_sample                    22.1 ms  22.2 ms    +0.5%     808.4 ms  838.5 ms   +3.7%
  lkj_cholesky_sample              5.3 ms   5.5 ms    +2.7%       1.18 s    1.16 s   -1.5%
  mixture_same_family_log_prob     2.1 ms   2.1 ms    +0.3%     102.1 ms  108.9 ms   +6.6%
  multivariate_normal_log_prob     282 µs   275 µs    -2.4%     164.2 ms  155.0 ms   -5.6%
  normal_log_prob                  749 µs   717 µs    -4.3%      57.4 ms   56.3 ms   -2.0%
  normal_sample                   22.9 ms  22.5 ms    -1.8%     219.3 ms  197.5 ms  -10.0%
  stick_breaking_transform         6.4 ms   6.5 ms    +0.9%     214.0 ms  216.5 ms   +1.1%
  student_t_log_prob               3.1 ms   3.0 ms    -1.3%      78.8 ms   79.1 ms   +0.4%
  truncated_normal_log_prob        730 µs   793 µs  (+8.7%)      56.2 ms   53.9 ms   -4.0%

handlers

                                  ──────── run time ───────     ─────── compile time ───────
  benchmark                       baseline   this PR      Δ     baseline   this PR         Δ
────────────────────────────────────────────────────────────────────────────────────────────
  initialize_model_hierarchical    40.3 ms   41.6 ms  +3.2%       3.89 s    4.12 s     +5.8%
  log_density_hierarchical          3.6 ms    3.7 ms  +1.2%       1.24 s    1.26 s     +1.5%
  nested_handler_stack              1.4 ms    1.4 ms  +4.5%       710 µs    529 µs  (-25.5%)
  potential_energy_and_grad          22 µs     21 µs  -1.8%     101.1 ms  101.5 ms     +0.4%
  predictive_forward_sampling     728.5 ms  720.0 ms  -1.2%     217.2 ms  221.7 ms     +2.1%
  trace_seeded_model                830 µs    828 µs  -0.3%     567.2 ms  554.2 ms     -2.3%

mcmc

                             ──────── run time ───────     ───── compile time ─────
  benchmark                  baseline   this PR      Δ     baseline  this PR      Δ
───────────────────────────────────────────────────────────────────────────────────
  hmc_logistic_regression    738.6 ms  749.7 ms  +1.5%       3.52 s   3.60 s  +2.5%
  nuts_dense_mass_funnel       1.19 s    1.18 s  -0.7%       2.77 s   2.78 s  +0.3%
  nuts_eight_schools           1.15 s    1.16 s  +0.9%       2.72 s   2.62 s  -3.7%
  nuts_hierarchical_glm        4.94 s    4.94 s  -0.1%       4.78 s   5.09 s  +6.4%
  nuts_logistic_regression     1.07 s    1.08 s  +1.5%       3.47 s   3.34 s  -3.7%
  nuts_vectorized_chains       2.56 s    2.50 s  -2.3%       2.84 s   2.98 s  +4.8%

svi

                                             ──────── run time ───────     ───── compile time ─────
  benchmark                                  baseline   this PR      Δ     baseline  this PR      Δ
───────────────────────────────────────────────────────────────────────────────────────────────────
  svi_autodelta_map_logistic                 311.7 ms  318.0 ms  +2.0%       3.23 s   3.39 s  +5.1%
  svi_autodiagonalnormal_hierarchical          1.03 s    1.05 s  +1.9%       5.06 s   5.36 s  +5.9%
  svi_automultivariatenormal_eight_schools   738.2 ms  748.7 ms  +1.4%       4.05 s   4.35 s  +7.3%
  svi_autonormal_logistic                    760.6 ms  775.4 ms  +1.9%       3.50 s   3.64 s  +4.1%
  svi_multi_particle_elbo                      1.48 s    1.51 s  +2.2%       3.66 s   3.77 s  +2.8%
  svi_trace_mean_field_elbo                    1.33 s    1.35 s  +1.7%       5.39 s   5.48 s  +1.6%

Methodology and environment

Each benchmark is set up untimed, then called once with the JAX caches cleared and several more times warm. Run is the fastest warm call; compile is the first call minus that, i.e. the tracing, lowering and XLA compilation the warm calls did not have to pay for.

Both refs were measured on the same runner over 2 interleaved round(s), taking the best observation per benchmark. A result is called neutral when it moves less than ±5% (run) or ±25% (compile), or when the measurement itself is under 1 ms (run) / 50 ms (compile) — a shared CI runner cannot resolve changes below that. Compile time gets the looser band because it is measured once per round rather than best-of-N, and swings by roughly 20% even between two runs of identical code. A delta shown in parentheses did clear its threshold, but on a measurement below the resolution floor, so it is reported without being called a change.

	baseline	this PR
ref	master	pr-cleanup-update
commit	5f9090f4	64bbac91
numpyro	0.21.0	0.21.0
jax	0.11.0	0.11.0
backend	cpu	cpu
python	3.14.7	3.14.7

Runner: Linux-6.17.0-1020-azure-x86_64-with-glibc2.39, 4 CPUs.

_{Produced by this benchmark run.}