fix(core): encode a float as its shortest decimal

A float was widened to double before encoding, so every float watch,
argument default and result read as the double nearest it: gravity
-9.81 showed as -9.8100004196167, a radius of 0.1 as 0.10000000149011612,
in `get`, `help` and call results alike. Agents read these, and copy
them back into `set` and `call`.

Codec<float> now encodes the shortest decimal that reads back as the
same float (std::to_chars), as a double. Decoding is unchanged, and so
is every value: 0.1 decodes to the float 0.1f came from. A double is
encoded as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

interfaces/Crafter.Debug.Core-Codec.cppm

@@ -192,7 +192,20 @@ export namespace Crafter::Debug {
 
     template<std::floating_point T>
     struct Codec<T> {
-        static Value Encode(T value) { return Value(static_cast<double>(value)); }
+        static Value Encode(T value) {
+            // A float as the shortest decimal that reads back as it: 0.1f
+            // widened is 0.10000000149011612, which is what an agent would
+            // otherwise read for every float.
+            if constexpr (sizeof(T) < sizeof(double)) {
+                if (std::isfinite(value)) {
+                    std::array<char, 32> text{};
+                    const auto written = std::to_chars(text.data(), text.data() + text.size(), value);
+                    double shortest = 0;
+                    if (written.ec == std::errc() && std::from_chars(text.data(), written.ptr, shortest).ec == std::errc()) return Value(shortest);
+                }
+            }
+            return Value(static_cast<double>(value));
+        }
 
         static void Decode(const Value& value, T& out, DecodeContext& context) {
             if (value.IsNumber()) {

tests/ShouldResolveWatchPaths/main.cpp

@@ -51,7 +51,7 @@ int main() {
     });
     Watch("score", [] { return 1234; });
 
-    Check(registry().Resolve("physics.gravity").value.AsNumber() == static_cast<double>(-9.81F), "a variable watch reads the variable");
+    Check(registry().Resolve("physics.gravity").value.AsNumber() == -9.81, "a variable watch reads the variable, a float as its shortest decimal");
     Check(registry().Resolve("bodies[1].pos.y").value.AsNumber() == 4, "a path reads into a watched value");
     Check(registry().Resolve("bodies[1]").value.Find("id")->AsInteger() == 2, "an index reads an item");
 

tests/ShouldRoundTripAggregates/main.cpp

@@ -6,7 +6,9 @@
 // maps, a scoped enum, Required) through Value and its text, with no
 // hand-written serialization. Fields the input leaves out keep their default
 // member initializers, null clears an optional, and an enum also accepts its
-// integer. Guards the reflection that command arguments and results rely on.
+// integer. A float encodes as its shortest decimal, 0.1 rather than the
+// 0.10000000149011612 it widens to, which every float watch used to show.
+// Guards the reflection that command arguments and results rely on.
 import std;
 import Crafter.Debug.Core;
 using namespace Crafter::Debug;
@@ -74,6 +76,13 @@ int main() {
     Check(!partial.target.has_value(), "null clears an optional");
     Check(partial.mode == Game::Mode::Build, "an enum accepts its integer");
 
+    Check(Encode(0.1F).Dump() == "0.1" && Encode(-9.81F).Dump() == "-9.81" && Encode(4.0F).Dump() == "4.0",
+          std::format("a float reads as its shortest decimal ({}, {})", Encode(0.1F).Dump(), Encode(-9.81F).Dump()));
+    float tenth = 0;
+    DecodeContext tenthContext;
+    Decode(Value::Parse(Encode(0.1F).Dump()), tenth, tenthContext);
+    Check(tenthContext.Ok() && tenth == 0.1F && Encode(0.1).Dump() == "0.1", "and decodes to the same float; a double is unchanged");
+
     float nan = 0;
     DecodeContext nanContext;
     Decode(Encode(std::numeric_limits<float>::quiet_NaN()), nan, nanContext);