pdftoraster: extract copy_image_rows() helper and add an ASan regression test

[rkt0209]

Follow-up to the review on #204: the standalone test-pdftoraster-copy-height.c re-implemented the copy loop, so it would pass even if the real write_page_image() regressed. This makes the test drive the real library code.

Refactor (no behaviour change): the copy loop in write_page_image() is extracted into a static void copy_image_rows(...) that write_page_image() now calls. The loop body is unchanged (existing filter tests still pass).

Test: test-pdftoraster-copy-height.c #includes the in-tree cupsfilters/pdftoraster.c to reach the static helper and its types, then calls copy_image_rows() under AddressSanitizer with a colordata buffer sized to exactly copy_height rows while cupsHeight is one row taller — the "rendered image shorter than page" case where the off-by-one lived. The buggy h <= copy_height reads one row past the buffer and ASan aborts; the fixed h < copy_height stays in bounds. Deterministic, no pdftoppm needed.

The .sh wrapper builds it under ASan and skips (Automake exit 77) when ASan is unavailable; the .c is listed in EXTRA_DIST so it ships in make dist.

Verified: fixed code PASS, buggy code FAIL (ASan heap-buffer-overflow in copy_image_rows), across system-2x / source-2.5.x / source-3.x.