alloc_cluster() just ignores read errors, trying next cluster until it
either succeeds (finds an empty one) or runs out of clusters (after
checking all of them). A large volume may have quite a lot of clusters -
eg. a 16 GB SD card with a standard format has about 2 million clusters.
When a "persistent" read error happens during the alloc_cluster() (after
a successful mount operation) - for example a volume is physically
disconnected in a very inconvenient moment or the volume is/gets damaged
and all further reads fail - then this loop becomes practically
infinite. In one application we found a damaged SD card, for which reads
of first 600-700 blocks work perfectly fine, but any read beyond that
results in a SDIO interface timing-out (the card will not switch to
expected state within specified time). As the timeout for the operation
is ~100 ms, then the function would loop for over 2 days. The same card
just fails to work in a PC, where any read beyond first ~350 kB (which
is about 700 blocks) fails with an I/O error.

Fix this by returning from alloc_cluster() with an error when any read
operation fails.

Fixes dlbeer#15