It is possible to trigger an integer overflow when parsing attribute
names that are longer than 2^31 bytes because we assign the result of
strlen(3P) to an `int` instead of to a `size_t`. This can lead to an
abort in vsnprintf(3P) with the following reproducer:

    blob=$(perl -e 'print "A " . "B"x2147483648 . "\n"' | git hash-object -w --stdin)
    git update-index --add --cacheinfo 100644,$blob,.gitattributes
    git check-attr --all path

    BUG: strbuf.c:400: your vsnprintf is broken (returned -1)

But furthermore, assuming that the attribute name is even longer than
that, it can cause us to silently truncate the attribute and thus lead
to wrong results.

Fix this integer overflow by using a `size_t` instead. This fixes the
silent truncation of attribute names, but it only partially fixes the
BUG we hit: even though the initial BUG is fixed, we can still hit a BUG
when parsing invalid attribute lines via `report_invalid_attr()`.

This is due to an underlying design issue in vsnprintf(3P) which only
knows to return an `int`, and thus it may always overflow with large
inputs. This issue is benign though: the worst that can happen is that
the error message is misreported to be either truncated or too long, but
due to the buffer being NUL terminated we wouldn't ever do an
out-of-bounds read here.

Reported-by: Markus Vervier <markus.vervier@x41-dsec.de>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>