fix(modbus): review r0 - FC23 opt-in, explicit maxReadGap 0 = off, legacy coil types load, one health writer

Coordinator rulings and Important findings from review r0:
1. FC23 is opt-in (new UseFC23ForMultiRegisterWrites, default false); a config
   without it - including the persisted "disableFC23": false every form-saved
   config carries - keeps the FC16 wire. DisableFC23 stays the hard kill-switch.
2. An explicit maxReadGap 0 means coalescing off again (its pre-lane meaning);
   the gap-8 default applies only to an absent key.
3. Coils/DiscreteInputs tags persisted with a register dataType (the old
   editor saved Int16) are coerced to Bool, never refused, with no deploy
   warning (a warning can be promoted to a blocking error). The editor loads
   them as Bool. Register region + Bool is still refused.
4. One comms-fault slot for reads and writes: a good read clears a write's
   transport fault and vice versa.
Minors: Dispose writes its own Unknown health (m1); persisted region/address
are authoritative over addressString so a Family change cannot remap a live
tag (m2); a malformed-but-delivered reply no longer fails the unit fast (m3);
a fresh socket published after a concurrent dispose is torn down (m4); the
editor resets address text per tag (m5); master merged (m6); the FC23 latch
log says re-init restores it (m7). Back-compat tests use the old persisted
JSON shapes and fail without the fixes.

Refs #741
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.Modbus.Contracts/ModbusDriverOptions.cs

@@ -89,16 +89,24 @@ public sealed class ModbusDriverOptions
     public bool UseFC16ForSingleRegisterWrites { get; init; } = false;
 
     /// <summary>
-    ///     Kill-switch for FC23 (Read/Write Multiple Registers). By default a multi-register holding
-    ///     write (2–121 registers: 32/64-bit values, arrays, strings) is sent as ONE FC23 that writes
-    ///     the block and reads the same block back, so the device confirms what it now holds. A unit
-    ///     that answers FC23 with exception 01 (MELSEC QJ71MT91, S7 Modbus servers) is moved to FC16
-    ///     automatically, for that write and every later one; one that fails FC23 at the transport is
-    ///     moved too (that one write is reported as a comms failure). Set <c>true</c> to always use
-    ///     FC16. Default <c>false</c>.
+    ///     Hard kill-switch for FC23 (Read/Write Multiple Registers): when <c>true</c> the driver never
+    ///     emits FC23, even with <see cref="UseFC23ForMultiRegisterWrites"/> set. Default <c>false</c>.
+    ///     FC23 is OPT-IN — a config without <see cref="UseFC23ForMultiRegisterWrites"/> keeps the FC16 wire
+    ///     behaviour, so the persisted <c>"disableFC23": false</c> every form-saved config carries changes
+    ///     nothing on the wire.
     /// </summary>
     public bool DisableFC23 { get; init; } = false;
 
+    /// <summary>
+    ///     Opt-in: send a multi-register holding write (2–121 registers: 32/64-bit values, arrays, strings)
+    ///     as ONE FC23 that writes the block and reads the same block back, so the device confirms what it
+    ///     now holds. A unit that answers FC23 with exception 01 (MELSEC QJ71MT91, S7 Modbus servers) moves
+    ///     to FC16 for that write and every later one until the next initialize; one that fails FC23 at the
+    ///     transport moves too (that one write is reported as a comms failure). Ignored when
+    ///     <see cref="DisableFC23"/> is set. Default <c>false</c> = FC16, the historical behaviour.
+    /// </summary>
+    public bool UseFC23ForMultiRegisterWrites { get; init; } = false;
+
     /// <summary>
     ///     Interval for the background re-probe loop that retries auto-prohibited
     ///     coalesced ranges. When non-null, every <c>AutoProhibitReprobeInterval</c>
@@ -110,14 +118,13 @@ public sealed class ModbusDriverOptions
     public TimeSpan? AutoProhibitReprobeInterval { get; init; } = null;
 
     /// <summary>
-    ///     Block-read coalescing budget. The read planner combines tags in the same (UnitId, Region)
-    ///     group whose addresses are at most this many registers (Coils / DiscreteInputs: bits) apart
-    ///     into a single FC03/FC04/FC01/FC02 read, then slices the response back into per-tag values.
-    ///     <c>0</c> coalesces only adjacent or overlapping tags (several <c>BitInRegister</c> tags on one
-    ///     status word still fold into one read). Default <c>8</c>. A range the device refuses is
-    ///     auto-prohibited and read per tag, so a protected hole costs one extra round trip, not data. To
-    ///     turn coalescing off entirely set <see cref="DisableReadCoalescing"/>, or exempt one tag with
-    ///     its <c>coalesceProhibited</c> key.
+    ///     Block-read coalescing budget. When non-zero, the read planner combines tags in the same
+    ///     (UnitId, Region) group whose addresses are at most this many registers (Coils / DiscreteInputs:
+    ///     bits) apart into a single FC03/FC04/FC01/FC02 read, then slices the response back into per-tag
+    ///     values (several <c>BitInRegister</c> tags on one status word fold into one read). <c>0</c> = no
+    ///     coalescing — an explicitly authored 0 keeps that historical meaning. Default <c>8</c> applies
+    ///     only when the key is ABSENT. A range the device refuses is auto-prohibited and read per tag.
+    ///     <see cref="DisableReadCoalescing"/> also turns it off; <c>coalesceProhibited</c> exempts one tag.
     /// </summary>
     public ushort MaxReadGap { get; init; } = DefaultMaxReadGap;