From reading the current code

it appears to me as if MPC-HC’s ISR defaults to using effectively no colour mangling (by using the video colourspace) if no header was set and doesn't explicitly implementYCbCr Matrix: None. None is a special header value to opt out of any colour mangling as far as possible for the renderer. Instead though all unknown values (so including None) are assigned ColorConvTable::NONE.
I'm not sure where this value is later used by ISR for ASS subs, but from the conditionals for ::NONE in PGSSub.cpp and DVBSub.cpp it appears like this amounts to a resolution dependent guess between TV.601 and TV.709.

Regardless of what ::NONE really does, this doesn't line up with xy-VSFilter/XySubFilter's behaviour and libass' recommendations as YCbCr Matrix: None should be different from an invalid value.
No header or an invalid value should be equal to an explicit TV.601 and while xy-VSFilter (but not XySubFilter) does a similar guessing for explicit None, it only does so due to lack of actual colour information which ISR appears to have. Assuming you want ISR to be more compatible with other renderers (and if I understood the code correctly), I suggest using yuvVideoMatrix for YCbCr Matrix: None and equate no or an invalid header to "TV.601".
For more information and some historic background, see e.g. libass docs:
https://github.com/libass/libass/blob/5f57443f1784434fe8961275da08be6d6febc688/libass/ass_types.h#L152-L199

Also the libass rendering backend should read track->YCbCrMatrix and deal with it as described in the above docs. It appears like this field is currently unused. (Same for ass_set_storage_size but that's unrelated to colour mangling.)