[SYCL] Ensure int-header fwd decls work with references

[erichkeane]

We weren't properly descending down into reference types, so we ended up
not forward declaring any types that were referred to only as a
reference type. This patch ensures we do so, just like we did with
pointer-type.

[premanandrao]

LGTM

[erichkeane]

@DoyleLi and @tfzhu and @bader : Looks like the Lit_With_Cuda bot is out of hard disk space again!

[AaronBallman]

LGTM!

[DoyleLi]

@DoyleLi and @tfzhu and @bader : Looks like the Lit_With_Cuda bot is out of hard disk space again!

It seems the system was used by someone and some configuration also changed. I have locked it.

[erichkeane]

3rd time is hopefuly the charm on Jenkins/Precommit! 1st two got hit with different flakey test failures.

[erichkeane]

@bader Another similar story.

1st failure was LLVM test suite WINDOWS, test was ESIMD/matrix_transpose.cpp
http://icl-jenkins.sc.intel.com:8080/blue/organizations/jenkins/SYCL_CI%2Fintel%2FPrecommit_Check_User/detail/Precommit_Check_User/8347/pipeline/

2nd failure was LLVM test suite LINUX, test was Reduction/reduction_nd_N_vars.cpp
http://icl-jenkins.sc.intel.com:8080/blue/organizations/jenkins/SYCL_CI%2Fintel%2FPrecommit_Check_User/detail/Precommit_Check_User/8389/pipeline/

3rd failure was LLVM test suite Windows again, test was Reduction/reduction_nd_N_vars.cpp (again, note windows vs linux)
http://icl-jenkins.sc.intel.com:8080/blue/organizations/jenkins/SYCL_CI%2Fintel%2FPrecommit_Check_User/detail/Precommit_Check_User/8398/pipeline/

This patch simply corrects a build error, so I don't think I could have caused this.

[bader]

@v-klochkov, @DenisBakhvalov, FYI. ESIMD and reduction tests seem to generate in a lot of false positives in pre-commit CI. It would be great if you can help to reduce the noise.

[DenisBakhvalov]

@v-klochkov, @DenisBakhvalov, FYI. ESIMD and reduction tests seem to generate in a lot of false positives in pre-commit CI. It would be great if you can help to reduce the noise.

ESIMD/matrix_transpose.cpp is fixed. If you rerun the CI, you should not see it again.