get_turnaround_flags(scanspeed method) fails for regular simulations

get_turnaround_flags(method="scanspeed") can fail with:
ValueError: cannot convert float NaN to integer

packages/sotodlib/tod_ops/flags.py", in 
get_turnaround_flags 
approx_samps_onescan = int(np.ptp(az) / approx_daz) 
 ValueError: cannot convert float NaN to integer

Cause: in line 220: approx_daz = np.median(daz[daz>np.percentile(daz, 95)])
https://github.com/simonsobs/sotodlib/blob/f915bfab48483efa7d794c9a139172a5b0977136/sotodlib/tod_ops/flags.py#L220

In simulations with constant scan speed, daz is discrete and often: p95 == max(daz)

So:

• daz > p95 --> empty array
• median --> NaN
• daz > p95 is too strict for regular simulations with stable scan speed where p95 == max(daz), leading to an empty selection and NaN; using >= avoids this.

I noted in simulations:
count(daz > p95) = 0
count(daz >= p95) > 0

Fix:
Use: approx_daz = np.median(daz[daz >= np.percentile(daz, 95)])

I checked that this solves the ValueError that often fails for my simulations. If this solution is accepted, I can open a PR.

Open sotodlib/tod_ops/flags.py and inspect get_turnaround_flags(method="scanspeed"), especially the approx_daz calculation around line 220. Reproduce the regular-simulation failure, then verify that the calculation no longer produces NaN and the ValueError is resolved.