tee-supplicant: renameat2() may not be supported

[etienne-lms]

Possible build issues using renameat2() since commit e3148e3 (#400) merged in OP-TEE release 4.8.0. The function was somewhat recently introduced and may not be supported.

Example of build error trace messages:

.../tee-supplicant/src/tee_supp_fs.c: In function 'ree_fs_new_rename':
.../tee-supplicant/src/tee_supp_fs.c:491:25: error: 'RENAME_NOREPLACE' undeclared (first use in this function)
  491 |                 flags = RENAME_NOREPLACE;
      |                         ^~~~~~~~~~~~~~~~

.../tee-supplicant/src/tee_supp_fs.c:493:13: error: implicit declaration of function 'renameat2'; did you mean 'renameat'? [-Werror=implicit-function-declaration]
  493 |         if (renameat2(tee_fs_fd, old_rel_filename, tee_fs_fd, new_rel_filename, flags)) {
      |             ^~~~~~~~~
      |             renameat

See for example these Buildroot CI build test reports (extracts from failures on target architectures):
x86_64: https://autobuild.buildroot.org/results/c5197057fbf8afb0a416e4dcd9e73ab82aa4b8b2/build-end.log
aarch64: https://autobuild.buildroot.org/results/60db3171134f732158cd0ceef4246788c972766e/build-end.log
sparc: https://autobuild.buildroot.org/results/f6f47442072ba415d363403ac38ca2ec17a55399/build-end.log

Should we provide a fallback to renameat() when RENAME_NOREPLACE is not defined by stdio.h?
That said, I understand that it could lead to race conditions as per e3148e3 commit message.

[jenswikl]

Would it make sense to add some dependencies for the optee_client package?