test: re-enable hipFFT and rocFFT callback tests

[mapatel-amd]

Summary

Re-enables the hipfft_callback and rocfft_callback tests by removing their
entries from .github/build_tools/skip_manifest.py.

Why they were skipped

rocFFT's default store callback was a static device function, so the compiler
appended a .static.<hash> suffix to its mangled name. HIP CLR could not resolve
that symbol through hipModuleGetGlobal, and both examples aborted at runtime:

:0:.../clr/hipamd/src/hip_global.cpp:209 : Cannot create GlobalVar Obj for symbol:
_ZL31store_cb_default_complex_double.static.4d22c59b5a98cebc

Tracked as
ROCm/rocm-systems#7263.

Why it is safe to re-enable

The issue was fixed by [ROCm/rocm-libraries@004a100615a](https://github.com/ROCm/r
ocm-libraries/commit/004a100615a) — "rocFFT: give default FFT callbacks external
linkage to fix hipModuleGlobal (#8477)", merged 2026-07-10; #7263 was closed
the same day.

The default callbacks are now declared with external linkage rather than static:

// projects/rocfft/library/src/device/kernels/callback.h:108
inline __device__ auto store_cb_default_complex_double =
store_cb_default<rocfft_complex<double>>;

No .static.<hash> suffix, so hipModuleGetGlobal resolves the symbol. The
commit is an ancestor of develop and predates the current nightly package by
roughly five weeks, so the fix is present in what CI tests against.

Changes

Removed both entries from SKIP_MANIFEST (15 lines, one file). The examples
themselves are unchanged.

Verification

Generator output for --channel nightly --target gfx1100 --distro ubuntu-24.04 --install-method tarball-multi-arch:

	skip_tests.txt	SKIP_FROM_TEST
before	hipfft_callback, rocfft_callback	`Libraries/hipFFT/callback
Libraries/rocFFT/callback`
after	(empty)	(empty)

All other manifest entries (rocDecode and the rest) are unaffected.

Empty-exclude-list edge case: these were the only entries carrying a ctest
key, so skip_tests.txt is now empty while the workflow still passes
--exclude-from-file unconditionally. Both halves check out:

• generate_skip_tests.py opens the file outside the if skip_tests: guard, so
  it is still created, just zero bytes.
• ctest --exclude-from-file <empty file> runs normally and exits 0 (verified on
  ctest 3.28.3).

Not verified on hardware

Every ROCm SDK available locally predates the fix (newest 7.15.0a20260702; the
fix landed 2026-07-10), so the examples could not be run directly. CI on the next
nightly is the real check.

If either test still aborts with the GlobalVar Obj error, the right response is
to reopen
ROCm/rocm-systems#7263 with
the new log rather than silently restoring the skip — the manifest entry would
otherwise hide a regression in a fix that is supposed to have shipped.

[regan-amd]

ROCm/rocm-libraries@004a100615a is expected to resolve the issue that motivated this test to be disabled in the first place, indeed.
However, the function-pointer callback feature that these samples illustrate are expected to fail on gfx1250 (known issue). Ideally, the re-enabled tests should be skipped/disabled/not tested on gfx1250, if possible.

[zichguan-amd]

We don't run gfx1250 in our CI workflows so if it works for other archs then it should be fine to reenable them.