Setting the number of tuning samples for custom compound steps

Questions v5
1

How can I set the number of tuning samples for custom compound steps?

Running this toy example

import pymc as pm
from pymc.step_methods.arraystep import BlockedStep
from pymc.util import get_value_vars_from_user_vars
from scipy.stats import norm

class CustomStep(BlockedStep):

    def __init__(self, vars, model = None):
        model = pm.modelcontext(model)
        self.vars = get_value_vars_from_user_vars(vars, model)

    def step(self, point: dict):
        new_point = point.copy()
        new_point[self.vars[0].name] = norm.rvs()
        stats = [{}]
        return new_point, stats  

with pm.Model():
    
   x = pm.Flat("x")
   custom_step = CustomStep([x])
   y = pm.Normal("y")

   trace = pm.sample(draws = 500, tune = 100, step = [custom_step])

prints

Multiprocess sampling (4 chains in 4 jobs)
CompoundStep
>CustomStep: [x]
>NUTS: [y]
 |████████████| 100.00% [2400/2400 00:01<00:00 Sampling 4 chains, 0 divergences]
Sampling 4 chains for 0 tune and 600 draw iterations (0 + 2_400 draws total) took 15 seconds.

indicating that tune is added to draws to yield the correct total number of samples (600); but this number is wrongly split into 0 tuning samples and 600 subsequent draws (instead of 100 and 500, respectively).
This also leads to the wrong number of draws in the trace object (e.g., trace.posterior.x.values.shape[1] == 600).