fix(ab): a refused operation is BadNotSupported, not a comms fault (Task 22 review r1; closes #571)

Native is the default, so a locally refused write must not look like a comms fault.

- AbLegacy: an AbTransportException with AbStatus.Unsupported (status-bit sub-element writes, PLC-5
  non-N reads, all PLC-5 writes; libplctag's ErrorUnsupported maps here too) now returns
  BadNotSupported on both the read and the write path (incl. the bit-RMW path). The tag handle is
  not evicted and driver health is not touched. A native runtime whose init read is refused locally
  is CACHED rather than disposed, so a poll of a refused tag does not re-create (and re-lease the
  pooled connection) every cycle.
- AbCip: the same mapping on the write path and both read paths (single + whole-UDT).
- Uses the existing named BadNotSupported constants (StatusCodeParityTests guard).
- Tests (falsified by disabling the mapping, and separately the keep-on-refusal): status-bit
  writes and PLC-5 read/write through the driver over the native runtime — BadNotSupported, zero
  wire requests, one create, health unchanged; AbCip refused read/write via a refusing fake plus a
  control that a non-refusal failure still evicts and degrades.
- Docs/CLAUDE.md: BadNotSupported; #571 known-gap wording removed.

Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe

docs/drivers/AbCip.md

@@ -159,6 +159,9 @@ The driver has two interchangeable wire transports, selected by one `DriverConfi
   alignment pad (84, not 82) (#570).
 - Status-bit sub-element writes (e.g. `T4:0.DN`) are Unsupported natively (they would clobber the control
   word); nothing goes on the wire. Sub-element arrays are Unsupported natively.
+- **A refused operation is not a comms fault.** Any operation a runtime refuses as Unsupported (in AbCip, e.g.
+  a native write whose type cannot be determined) reaches the client as `BadNotSupported`, keeps the tag
+  handle, and leaves driver health unchanged; only real transport failures evict and degrade.
 
 ## Known gaps