feat(chat): surface per-model reasoning capabilities (#946) (#951)

Add ModelCapabilities (capabilities: Map<String, ModelCapabilities>?)
to ModelProvider matching backend /api/model/options shape
{model: {fast, reasoning, can_disable_reasoning}}. Nullable/defaulted
for backward compat with old backends (ignoreUnknownKeys+SnakeCase).

Chat composer: gate reasoning dropdown on can_disable_reasoning and
supportsReasoning — "None" disabled with "reasoning always on" hint
when can_disable==false, chip disabled with "No reasoning" when
reasoning==false. Missing keys offer full scale.

Sync currentModelCapabilities from cachedModelOptions on
SessionInfo, resume, preload/refresh, and optimistic /model swap;
clear on resetSessionState. Used by ComposerToolbar via ChatScreen.

Model picker: surface same hints in ModelPickerDialog (pinned+all)
and ModelScreen ProviderCard.

Verified: ktlintCheck OK, compileDebugKotlin OK (rerun-tasks).

Co-authored-by: M57 <hy4ri@users.noreply.github.com>

app/src/main/java/com/m57/hermescontrol/data/model/ModelOption.kt

@@ -18,6 +18,14 @@ data class ModelProvider(
     val authenticated: Boolean? = null,
     val auth_type: String? = null,
     val warning: String? = null,
+    val capabilities: Map<String, ModelCapabilities>? = null,
+)
+
+@Serializable
+data class ModelCapabilities(
+    val fast: Boolean? = null,
+    val reasoning: Boolean? = null,
+    val can_disable_reasoning: Boolean? = null,
 )
 
 @Serializable