fixed javadoc error

Author: RealAntEngineer

src/main/java/com/rae/formicapi/foundation/math/operators/backend/gpu/PortableKernel.java

@@ -32,7 +32,7 @@
  * executor it was bound to last. Native kernels are cached per executor, so
  * re-binding to an executor already seen is a cheap lookup, not a recompile.
  *
- * <h1>GLSL conventions (Vulkan)</h1>
+ * <h2>GLSL conventions (Vulkan)</h2>
  * Write only the body of the shader -- no {@code #version}, no
  * {@code local_size}, no {@code #extension}; the Vulkan backend adds those
  * (extensions are derived from the feature names given to {@link #glsl}).

src/main/java/com/rae/formicapi/foundation/math/operators/backend/gpu/vulkan/VulkanGpuExecutor.java

@@ -30,8 +30,7 @@
  * fills, copies and compute dispatch. Like {@code OpenCLGpuExecutor} it owns
  * no kernel source -- those live as {@code PortableKernel} constants next to
  * the {@code Gpu*Vector}/{@code Gpu*Matrix} classes that use them.
- *
- * <h1>Execution model</h1>
+ * <h2>Execution model</h2>
  * <ul>
  *   <li><b>Serials and fences.</b> Every operation is its own one-time
  *       command buffer submitted to one queue with its own fence, and gets the
@@ -64,7 +63,7 @@
  *       and bookkeeping happen under one monitor; host waits happen outside it.</li>
  * </ul>
  *
- * <h3>Requirements and limits</h3>
+ * <h2>Requirements and limits</h2>
  * Vulkan 1.2. {@code shaderFloat64} is <em>optional</em>: devices without
  * it work for the float/int API, and only double-precision kernels are
  * refused (see {@link #supportsFeature}); the double helpers throw