Increase width of padded output in stdio-common/tst-put-error

[Roland McGrath]

> I don't know why the test uses the present approach with setlinebuf.  (To 
> some extent I think this test is illustrating another development 
> guideline: adding new tests to an old testcase is often a bad idea as it 
> can cause what appear to be regressions but aren't.  Any reasons for how 
> things are done in the old (2005) part of the test may not apply to the 
> bit added last month about testing padded printing.)

Sounds like the right fix is to revert last month's addition and instead
add a new test, which can use an explicit buffer size to be a robust test
of what it intends to test regardless of the details of the environment.
Or you could just wait until someone makes a version of tmpfs that uses
megabytes or more for st_blksize and it breaks again.