examples/c: rebuild BPF objects on header changes in CMake

[Dhanya16]

Problem: bpf_object() supports optional headers in DEPENDS, but examples/c/CMakeLists.txt only passed the .bpf.c source. Touching a shared header (e.g. bootstrap.h) rebuilt userspace via CMake's normal C dependency scanning, but not the BPF object or skeleton.

Change: Pass ${app_stem}.h into bpf_object() when that file exists.

Follow-up to #371 (CMake path noted by @danielocfb ).

Before:

$ stat -c '%y %n' bootstrap.bpf.o
2026-08-31 15:06:42.837222615 +0530 bootstrap.bpf.o
$ touch ../bootstrap.h
$ cmake --build . --target bootstrap 2>&1 | grep -E 'Building BPF|skel|bootstrap\.c'
[ 85%] Building C object CMakeFiles/bootstrap.dir/bootstrap.c.o
$ stat -c '%y %n' bootstrap.bpf.o
2026-08-31 15:06:42.837222615 +0530 bootstrap.bpf.o

After:

$ stat -c '%y %n' bootstrap.bpf.o
2026-08-31 16:03:20.703857431 +0530 bootstrap.bpf.o
$ touch ../bootstrap.h
$ cmake --build . --target bootstrap 2>&1 | grep -E 'Building BPF|skel|bootstrap\.c'
[ 85%] [clang] Building BPF object: bootstrap
[ 90%] [skel]  Building BPF skeleton: bootstrap
[ 95%] Building C object CMakeFiles/bootstrap.dir/bootstrap.c.o
$ stat -c '%y %n' bootstrap.bpf.o
2026-08-31 16:04:14.114486381 +0530 bootstrap.bpf.o

[danielocfb]

Not entirely convinced this is the entire story, as I think it relies on a bug in the existing cmake plumbing. But it's probably better than the status quo.