WIP: firmware: guard I2C transactions with timeouts. If we don't add timeouts on I2C transactions, then the USB request that caused it (for reads) or the next USB control request (for writes) will silently hang forever until the device is power cycled. This is clearly undesirable. However, this does not happen at all during normal operation, and although it hardens the device, the FX2 firmware size comes at a premium, and blindly spraying these guard conditions adds a nontrivial increase. Therefore this code, although tested, is not currently used.

applet.i2c_master: require revC+. The I2CPadsWrapper hack was horrible to use even when it worked. I can't in good conscience recommenda anyone with a revA/B do that, so let's get rid of it.

device.hardware: add pull set support.

firmware: add pull get/set support.

firmware: persist voltage limit on writes, not reads. Oops... this would have written the EEPROM on each `glasgow voltage` invocation.

applet: add required_revision field.

applet.jtag: enumerate-ir: make more robust to missing DRs. A missing DR is a DR that leaves TDO unconnected when it is selected.

applet.jtag_mips: update the lists of tested configurations.

applet.jtag: enumerate-ir: handle incorrect TAP selection.

applet.jtag: reset TAP chain before selecting a TAP. Selecting a TAP requires reading IDCODE/BYPASS from DR, which means that a TAP chain reset was a precondition for select_tap(), but this wasn't intended or documented. The current design requires a reset, so just add that. Since all derived applets want to select a TAP, remove the TAP reset via TMS from run(); however, keep TAP reset via TRST#, because that enables the TRST# driver.