flag conflict `-Ddynamic=false` and `-Dstatic_planning`

[MirkoCalvi]

Description

For a correct codegeneration when you set -Dstatic_planning you should also set -Ddynamic=false, one exclude the other. I would consider to eliminate -Dstatic_planning and use the static tensor allocation as default method, and when -Ddynamic=false is set we use the dynamic allocation.

Pay also attention to TODO at line 106 in src/codegen/cg_v1/predict/predict.zig reported here:

    // Allocate output tensors for dynamic mode (only when NOT using plan-based execution)
    if (codegen_options.dynamic) {
        // TODO: When using plan-based execution, the PlanEmitter handles all tensor allocation
        // For now, we skip traditional output tensor allocation to avoid duplicates
        // In a complete implementation, we'd check if plan-based execution is enabled

        // Temporarily skip output tensor allocation to avoid conflicts with PlanEmitter
        // const output_tensors: []TensorZant = try IR_utils.getOutputs(tensorZantMap);
        // for (output_tensors) |*tz| {
        //     _ = try emit.ShapeEmitter.emit(writer, tz);
        //     const sanitized_name = try tz.getNameSanitized();
        //     const type_str = tz.ty.toString();
        //     try writer.print("    var tensor_{s} = Tensor({s}).fromShape(&allocator, &shape_tensor_{s}) catch return {d};\n", .{ sanitized_name, type_str, sanitized_name, templates.RC.INIT_ERROR });
        //     //since we are using dynamic inference  we also have to free the output_tensor so to avoid leaks, seee how I return the output tensor in writeReturn()
        //     try writer.print("    defer tensor_{s}.deinit();", .{sanitized_name});
        // }
    }

Update all the DOCS by consequence