…e_api_test

* origin/sycl: (1188 commits)
  [SYCL][CUDA] Improve CUDA backend documentation (intel#1293)
  [SYCL] Emit textual IR when -S -fsycl-device-only is used (intel#1314)
  [SYCL] Add prototype of group algorithms (intel#1236)
  [SYCL] XFAIL test on windows to unblock pulldown
  Allow Intel Loop Controls only with SPV_INTEL_fpga_loop_controls
  Apply suggested assert msg change
  Implement SPV_INTEL_io_pipes extension
  [SYCL] Fix dependencies for SYCLLowerIR (intel#1321)
  [CI] Allow builds without pre-downloaded OpenCL in configure.py (intel#1317)
  [SYCL] Move SYCL headers from standard clang location (intel#1308)
  [mlir] Add support for generating dialect declarations via tablegen.
  Be more strict when checking existence of foo
  [CodeGenPrepare] Freeze condition when transforming select to br
  [ORC] Remove an undefined static method from LLJIT.
  [JITLink][AArch64] Fix incorrect capitalization in a testcase name.
  [ORC] Print symbol flags and materializer name in ExecutionSession::dump.
  [JITLink][MachO] Re-apply b64afad, MachO linker-private support, with fixes.
  Basic Block Sections Support.
  Test commit.
  [SYCL][Doc] Deploy documentation for PI (intel#1318)
  ...

intel#1321)

I had a bit of trouble working out why lit was failing to start after
configuration had succceeded. Turns out I was passing a partial tuple
to cmake `-DCHECK_SYCL_ALL` due to a typo in a shell variable.

This patch verifies that both parts of each tuple in the
`CHECK_SYCL_ALL` variable are nonempty and terminates configuration with
an error message otherwise, which should make it a bit more obvious
what's happening.