I'm not sure we've gotten anywhere on this design...
How about we keep it simple and effective?
I think the simplest of all solutions would be to place the volume pot for the SN76489 at the backplate of the card, so that you can easily turn it off during bootup, and dial it up once you have booted up and silenced the chip with a simple utility.
For booters, I think you'd be okay if you would boot the system first, silence the card, and then do a warm boot (ctrl-alt-del) to run the booter game. It shouldn't reset the SN76489, and therefore it should remain silent.
If that fails, I could build a simple utility (if one does not already exist) to launch a booter from DOS.
A variation on that theme could be to have a simple switch at the back of the card as well, to silence the output without having to adjust the volume. The switch could either just break the gnd on the output jack, or switch the gnd for the op-amp. Whichever way is easiest to stop it outputting any signal.
Since both Tandy and PCjr explicitly silence all 4 channels from the BIOS, I think it will be wrong to assume that games will silence all channels themselves at bootup, so whenever a game doesn't use all 4 channels, the unused channels might continue to output the 'startup' tone indefinitely.
And it seems that silencing the chip automatically is blowing the complexity up by a few levels, so that's probably not feasible.
I wonder though... how difficult would it be to write 4 bytes to the card with discrete logic?
Because basically we only need to do the following sequence:
9F, BF, DF, FF
So the following pattern is always static:
9F -> 10011111
We could hardwire those.
Only bits 5 and 6 need to change during this pattern:
00, 01, 10, 11
So technically we'd just need to send a sequence of 4 2-bit values to the card, with the handshaking required to the SN76489.
An extension rom or microcontroller is massive overkill of course. But how difficult would it be to do this in discrete logic? It'd be like a 2-bit adder, where the input needs to be triggered by the SN76489 lines I suppose. Don't have the details worked out yet.
I think the main problem for either a discrete circuit or a microcontroller is to somehow mulitplex between the ISA port and our 'alternative' data source.
I can think of using some and-gates, so that the ISA port signals can only come through if another signal is also active at the same time.
That would allow you to put the ISA bus on 'ignore' during initialization, and then make the ISA bus active at the end.
But it'd still blow up the complexity considerably. You'd need 10(?) extra AND gates to pull that off.