Fix invalid mapping to zero page caused by off-by-one bug

The `zero_end` bound is exclusive, but we treat the `end_page` as inclusive. So when `zero_end` is page-aligned, we allocate one additional bss page. If this page was already mapped to some other segment, we remap it to a page with random content.

This is the same bug as #362.

src/page_table.rs

@@ -187,7 +187,7 @@ pub(crate) fn map_segment(
                     zero_start.as_u64(),
                     Size4KiB::SIZE,
                 )));
-                let end_page = Page::containing_address(zero_end);
+                let end_page = Page::containing_address(zero_end - 1usize);
                 for page in Page::range_inclusive(start_page, end_page) {
                     let frame = frame_allocator
                         .allocate_frame(MemoryRegionType::Kernel)