Code review comment for lp://qastaging/~alan-griffiths/mir/fix-1483223

The best news is input-reader-perf.bin is 10MB smaller.

And libmirserver is 6MB smaller, but libmircommon+libmirclient are now bigger to compensate. So the overall gain is almost negligible.

(1) Shouldn't MIR_CLIENT_DETAIL_9 become MIR_CLIENT_DETAIL_9.1 for new functions? Even as an internal ABI I feel we should probably manage it like the others.

(2) Doesn't putting the input code into libmircommon make more sense than libmirclient? I gather that would require moving some exception code mentioned in the bug. But it feels like we're using libmirclient as a helper library for libmirserver, which seems ugly. This isn't a new problem; libmirserver is already linking to libmirclient to support nesting (I assume), but I thought that was something we'd like to move away from eventually.