webfs: reset readline Ibuf rp/wp to buf start when buffer drains

readline() in sys/src/cmd/webfs/buf.c reads response bytes
into the 4096-byte b->buf via ioread(..., b->wp,
sizeof(b->buf)/2).  When b->rp catches up to b->wp, readline
treats the buffer as empty and issues another ioread but
never resets rp/wp.  Over many drain/refill cycles the
pointers walk toward the end of b->buf, and the next ioread
overflows into the adjacent rp/wp fields of the same Ibuf
struct.  The next *b->rp++ dereferences the corrupted
pointer and webfs faults inside readline.

Any webfs consumer driving sustained multi-connection reads
hangs on heavy pages.  hget (single-connection, sequential)
does not cycle the buffer enough to trip it on the same URL.

Fix: reset rp and wp to &b->buf[0] each time the buffer is
consumed to empty.  Single-line change.  No behaviour change
for any input; the buffer was already semantically empty,
and initibuf / unreadline already reset rp/wp in their own
paths — this closes the third drain path (readline itself).

RFC relevance: none

--- sys/src/cmd/webfs/buf.c
+++ sys/src/cmd/webfs/buf.c
@@ -57,6 +57,7 @@

 	for(p = buf;;){
 		if(b->rp >= b->wp){
+			b->rp = b->wp = b->buf;
 			n = ioread(b->io, b->fd, b->wp, sizeof(b->buf)/2);
 			if(n < 0)
 				return -1;