applet.interface.jtag_probe: completely rework error handling. JTAGProbeInterface historically suffered from underuse of exceptions; it was originally written to return void values on error paths since it strictly validates all input at every opportunity, and this does not survive collision with the real world. (Can you believe there are production JTAG devices where IDCODE is 32 zeroes?) Scanning, as well as any future manual overrides, have to relax validation greatly. Unfortunately, this approach caused proliferation of identical error handling code in every caller, error propagation boilerplate, and polymorphic return values. It was awful. This commit changes all remaining code to use exceptions and, in most cases, accept a check= argument to opt into returning void values where truly necessary. The check= argument is not forwarded; callers that need to customize error handling in a fine-grained way should inline and tweak the code that needs to behave differently.

Fix tests broken in commit 77425539.

applet.interface.jtag_probe: fix race condition in scan_xr().

applet.interface.jtag_probe: don't track IR offsets. NFCI. Per-TAP IR offsets are redundant, pointless, and complicate the logic of specifying IR lengths manually.

applet.interface.jtag_probe: improve TAP selection logic. * Require --tap-index option if more than one TAP is found. Before this commit, TAP #0 was implicitly chosen. * Extract redundant error handling code into select_tap(). * Convert logger errors into exceptions where appropriate.

applet.interface.jtag_probe: remove read_dr(idempotent=) argument. Before this commit, this unused option had to be threaded through a lot of unrelated code, which polluted interfaces. Moreover, since commit 42144aa6, it is also worse than useless, because scan_dr() does the exact same thing more safely (it does not rely on provided DR length).

applet.interface.jtag_probe: remove scan_dr(zero_ok=) argument. Before this commit, this option used only in enumerate-ir had to be threaded through a lot of unrelated code, which polluted interfaces and made the logic hard to understand. Also, tighten the post-conditions on scan_xr_length() such that it always requires a valid scan result (finite non-zero length).

applet.interface.jtag_probe: move XR scan limits to JTAGProbeInterface. Before this commit, these rarely useful options had to be threaded through a lot of unrelated code, which polluted interfaces and was done quite inconsistently. Also, do not call add_run_arguments() from add_run_tap_arguments(). This was inconsistent with our conventions and required an `access` argument in add_run_tap_arguments() that was otherwise unnecessary.

applet.interface.jtag_probe: reliably detect overlong XR scans.

applet.interface.jtag_probe: clarify enumerate-ir documentation. Also, lower log level for DR[?] to WARN since it is as noncompliant as DR[0] and neither of these results usually indicates a problem.