This should have some comment that explains what the function does/returns.

I know you just copied the API from what there already was, but don't think it's a nice API to have the function return three things (error status, length in frames, and start frame) encoded as two values, which are then split up as one return value and one return parameter. Something like this would be much nicer IMO:

bool BKE_cachefile_get_sequence_len(const char *filename, int *r_start_frame, int *r_sequence_length_frames);

This is just signf(frame_a->framenr, frame_b->framenr).

This entire block should be moved to some Alembic-specific function, and not be in the generic cachefile function.

These don't seem to be in use.

Include units, as offset = 0 doesn't mean anything.

Flip the condition and put the error handling with the error check. When moving things into their own function, you can just do a return from the error condition:

if (sequence_len < 0) {
  WM_report(RPT_ERROR, "Could not determine the time range of the Alembic archive.");
  return;
}

This probably should also have a poll function, to see if there is even a filename set.

Nice!

This could be a BLI_assert_msg(time_sampling_ptr, "could not get Alembic time sampling even though it was declared as available") or something along those lines.

Why should the default time sampling be avoided? This is not clear from the code.

This should have a comment to explain what it's for.

It should also explain which side of the interval is open, i.e. does this encode [min_time, max_time] or [min_time, max_time) or any other combo of open/closed sides?

Why is min_time == max_time invalid?

This should probably be max_time != -std::numeric_limits<Alembic::Abc::chrono_t>::min();, but in that case min_time < max_time is guaranteed to return false anway. Is this check really required?

IMO this could be shortened to

if (!archive || !archive->valid()) {
  return false;
}

Just personal opinion, though. Feel free to ignore.