So, since the sender for an async serial interface like UART doesn't send any sort of clock/strobe to the receiver, I assume I will either have to generate a clock based on the MCU's system clock via timer interrupts out of GPIO pins, or use a separate external 3.58Mz oscillator?

If a separate clock is used, wouldn't there potentially be a small margin of error if the clocks/baudrate of the UART and the smart card aren't exactly synced? Also, what then happens to ISO7816's capabilities of changing the baudrate? Can I just fix it to 9600baud if I don't need any more speed than that?

I've actually downloaded the PDF article, it appears to be a free download. I've read it, and the author does indeed use an external oscillator and wired the TX/RX together with a 10k pullup. He also didn't implement any error detection, including parity check, which probably makes things a lot easier.