Fine performance metrics dashboard crashes w/ spill activity

[crusaderky]

This code is sized so that it triggers spill/unspill activity:

import dask.array as da
import distributed

client = distributed.Client(n_workers=4, threads_per_worker=1, memory_limit="2 GiB")
a = da.random.random((14_000, 14_000))
b = (a @ a.T).sum()
b.compute()

If I open the Fine Performance Metrics dashboard after I run it, I get a blank page with 500 Internal server error, while on the stderr I get

BokehUserWarning: ColumnDataSource's columns must be of the same length. Current lengths: ('activity', 11), ('angle', 11), ('color', 0), ('text', 11), ('value', 11)
2023-06-01 15:33:53,229 - bokeh.core.property.validation - ERROR - Keyword argument sequences for broadcasting must all be the same lengths. Got lengths: [0, 11]
Traceback (most recent call last):
  File "/home/crusaderky/github/distributed/distributed/utils.py", line 760, in wrapper
    return func(*args, **kwargs)
  File "/home/crusaderky/github/distributed/distributed/dashboard/components/scheduler.py", line 3510, in update
    task_exec_barchart = self._build_task_execution_by_prefix_chart(
  File "/home/crusaderky/github/distributed/distributed/dashboard/components/scheduler.py", line 3598, in _build_task_execution_by_prefix_chart
    renderers = barchart.vbar_stack(
  File "/home/crusaderky/miniconda3/envs/distributed39/lib/python3.9/site-packages/bokeh/plotting/figure.py", line 588, in vbar_stack
    for kw in double_stack(stackers, "bottom", "top", **kw):
  File "/home/crusaderky/miniconda3/envs/distributed39/lib/python3.9/site-packages/bokeh/plotting/_stack.py", line 83, in double_stack
    raise ValueError("Keyword argument sequences for broadcasting must all be the same lengths. Got lengths: %r" % sorted(list(lengths)))
ValueError: Keyword argument sequences for broadcasting must all be the same lengths. Got lengths: [0, 11]

I reproduced the above both with bokeh 2 and bokeh 3.

The problem disappears if I size either my problem or the cluster so that spill/unspill events are never generated.
I'm attaching a trimmed-down dump of the metrics, which will allow you to reproduce it deterministically. Just copy-paste the below in a jupyter notebook and open the fine performance metrics dashboard page:

import distributed

client = distributed.Client()
metrics = {
    ("execute", "chunk_sum-aggregate-sum", "disk-read", "seconds"): 1.0051379720016484,
    ("execute", "chunk_sum-aggregate-sum", "disk-read", "count"): 16.0,
    ("execute", "chunk_sum-aggregate-sum", "disk-read", "bytes"): 2059931767.0,
    ("execute", "chunk_sum-aggregate-sum", "disk-write", "seconds"): 0.1692888050001784,
    ("execute", "chunk_sum-aggregate-sum", "disk-write", "count"): 2.0,
    ("execute", "chunk_sum-aggregate-sum", "disk-write", "bytes"): 268435938.0,
}
client.cluster.scheduler.cumulative_worker_metrics.clear()
client.cluster.scheduler.cumulative_worker_metrics.update(metrics)