⚡️ Throttle `aCompCor_cosine_filter`

[shnizzedy]

User interface changes

This node is using way more than its allocated memory again, this time with the fmriprep-options preconfig on some HCP data.

Until #1166 / #1404 / #1480 / #1505 / #1506 / #1509 are resolved, we'll allocate all available memory for each aCompCor_cosine_filter thread so that such nodes will necessarily run one-threaded and won't overrun memory as long as the total allocation is sufficient

Project sub-parts and estimates

1. Enable throttling ①
2. Set throttling on aCompCor_cosine_filter ①
3. Test throttled nodes ⑤

Timeline notes

No response

Timesink notes

No response

Frustration notes

No response

Design decisions and notes

• Throttle all cosine_filters or just aCompCor?
• Possibly also throttle calc_ants_warp ― it's been consistently problematic with overrunning its memory allocation (Calc Ants Warp Crashing #1308)
• Throttle cal_DVARS? (Pipelines hanging on DVARS calculation #1301)