t/helper: improve "genrandom" test helper

t/helper: improve "genrandom" test helper

The `test-tool genrandom` test helper can be used to generate random
data, either as an infinite stream or with a specified number of bytes.
The way we handle parsing the number of bytes is lacking though:

  - We don't have good error handling, so if the caller for example uses
    `test-tool genrandom 200xyz` then we'll end up generating 200 bytes
    of random data successfully.

  - Many callers want to generate e.g. 1 kilobyte or megabyte of data,
    but they have to either use unwieldy numbers like 1048576, or they
    have to precompute them.

Fix both of these issues by using `git_parse_ulong()` to parse the
argument. This function has better error handling, and it knows to
handle unit suffixes.

Adapt a couple of our tests to use suffixes instead of manual
computations.

Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>