sparse test: fix for 64k page size

[Fabian-Gruenbichler]

this test failed on Debian's CI on ppc64el, because the page size is 64k there, which means the block size on tmpfs is as well, so attempting to create 32k big holes cannot work.

[Fabian-Gruenbichler]

alternatively, the block size here could be #[cfg()]-ed based on the target arch

[Fabian-Gruenbichler]

link to the failing CI test: https://ci.debian.net/packages/r/rust-tar/testing/ppc64el/54192787/#S9

[Fabian-Gruenbichler]

I actually only realized afterwards there's another test case that has an assert that fails on powerpc64 for the same reason..

on the Debian side I just skip that assert for now, but the whole test would need a bigger refactor to handle different sparse granularities properly..

Index: tar/tests/all.rs
===================================================================
--- tar.orig/tests/all.rs
+++ tar/tests/all.rs
@@ -1301,7 +1301,8 @@ fn writing_sparse() {
     let data = t!(ar.into_inner());
 
     // Without sparse support, the size of the tarball exceed 1MiB.
-    #[cfg(target_os = "linux")]
+    // ppc64el has 64k pages and thus 64k blocks on tmpfs
+    #[cfg(all(target_os = "linux", not(target_arch = "powerpc64")))]
     assert!(data.len() <= 37 * 1024); // ext4 (defaults to 4k block size)
     #[cfg(target_os = "freebsd")]
     assert!(data.len() <= 273 * 1024); // UFS (defaults to 32k block size, last block isn't a hole)

would a patch adapting that properly also be welcome?