Add missing includes

[thilinarmtb]

Description

Currently, when we run OCCA tests, they use the automatically generated *_codegen
files from the build directory. This is because, we do the following two things:

• -I${OCCA_INSTALL_DIR} is added to the compile command in device::buildKernel()
  methods.
• If OCCA_INSTALL_DIR is not defined, it defaults to OCCA_BUILD_DIR.

Since we never set OCCA_INSTALL_DIR, we have always been using *_codegen files from
the build directory. When I tried running the tests by setting OCCA_INSTALL_DIR, I get
compilation errors since we don't have the required *_codegen files at the correct places
in the install directory. This PR fixes this issue.

PS: I only tested the CMake build for now. May need fixes for the GNU Make build.

Todo

• Run the GitHub CI tests with OCCA_INSTALL_DIR set.

[thilinarmtb]

I tested this on both Polaris and Aurora with the following changes.

On Aurora:

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 51597010..4efe545b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -41,7 +41,7 @@ option(OCCA_ENABLE_METAL  "Build with Metal if available" ON)
 option(OCCA_ENABLE_DPCPP "Build with SYCL/DPCPP if available" ON)

 option(OCCA_ENABLE_TESTS    "Build tests"               OFF)
-option(OCCA_ENABLE_EXAMPLES "Build simple examples"     OFF)
+option(OCCA_ENABLE_EXAMPLES "Build simple examples"     ON)
 option(OCCA_ENABLE_FORTRAN  "Enable Fortran interface"  OFF)
 option(OCCA_CLANG_BASED_TRANSPILER      "Build with occa-transpiler dependecy" OFF)

diff --git a/examples/cpp/01_add_vectors/main.cpp b/examples/cpp/01_add_vectors/main.cpp
index 66e449bb..bf46c38c 100644
--- a/examples/cpp/01_add_vectors/main.cpp
+++ b/examples/cpp/01_add_vectors/main.cpp
@@ -31,6 +31,11 @@ int main(int argc, const char **argv) {

   //---[ Device Setup ]-------------------------------------
   device.setup((std::string) args["options/device"]);
+  device.setup({
+    {"mode"     , "DPCPP"},
+    {"platform_id", 0},
+    {"device_id", 0},
+  });

   /*
    * Examples of setting up a device in other backend modes:

On Polaris:

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 51597010..4efe545b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -41,7 +41,7 @@ option(OCCA_ENABLE_METAL  "Build with Metal if available" ON)
 option(OCCA_ENABLE_DPCPP "Build with SYCL/DPCPP if available" ON)

 option(OCCA_ENABLE_TESTS    "Build tests"               OFF)
-option(OCCA_ENABLE_EXAMPLES "Build simple examples"     OFF)
+option(OCCA_ENABLE_EXAMPLES "Build simple examples"     ON)
 option(OCCA_ENABLE_FORTRAN  "Enable Fortran interface"  OFF)
 option(OCCA_CLANG_BASED_TRANSPILER      "Build with occa-transpiler dependecy" OFF)

diff --git a/examples/cpp/01_add_vectors/main.cpp b/examples/cpp/01_add_vectors/main.cpp
index 66e449bb..b37ff8d8 100644
--- a/examples/cpp/01_add_vectors/main.cpp
+++ b/examples/cpp/01_add_vectors/main.cpp
@@ -31,6 +31,10 @@ int main(int argc, const char **argv) {

   //---[ Device Setup ]-------------------------------------
   device.setup((std::string) args["options/device"]);
+  device.setup({
+    {"mode"     , "CUDA"},
+    {"device_id", 0},
+  });

Then run the 01_add_vectors example using the following environment:

export OCCA_INSTALL_DIR=<occa_install_dir>
./examples_cpp_add_vectors

The development branch fails in this case and this PR works fine.