Fix null pointer dereference on SPIR-V assembly failure

[esoalba]

Fixes #1597

shaderc_assemble_into_spv() leaves output_data_ null when assembly fails, but GetBytes() unconditionally dereferences it, crashing glslc, the C API (shaderc_result_get_bytes), and the C++ API (cbegin()/cend()) on any invalid .spvasm input.

This adds a null check before the dereference. Verified: invalid assembly now errors gracefully (exit 1 with an assembly error message) instead of segfaulting; valid assembly still compiles; C API returns nullptr on failure.

Unit test added (3cbbe7a): AssembleStringTest.FailedAssemblyGetBytesIsNull asserts that a failed assembly result has null bytes and zero length. Without the fix the test crashes with a null pointer dereference (SIGSEGV); with the fix it passes. Full test suite: 151 tests in shaderc_test, 126 in shaderc_cpp_test, all passing.

Reported via Google OSS VRP (g.co/vulnz).

[dneto0]

This should get a unit test also.

[dneto0]

This should get a unit test also.

I've added unit tests. And rebased on top of #1603

[esoalba]

Thanks @dneto0 — appreciate the review, and for taking the fix further. Hardening all the result getters against a null result object plus the NullResultObj and GetBytes* tests is a more complete fix than my original change, and the rebase on #1603 is appreciated.
On the test request: I had a complementary end-to-end regression test ready to push — shaderc_assemble_into_spv on invalid input, then asserting shaderc_result_get_bytes() returns nullptr on the failed result (verified it SIGSEGVs without the null check). I held off since your tests already cover the null-safety at the unit level. Happy to add it as an integration-level check if you think it's worth having; otherwise I'll leave the branch as-is.