wc: derive column width from the size of stdin

[Socialpranker]

wc pads its counts to a width derived from the total size of its inputs, but
stdin was always treated as a stream of unknown length, even when it is a
regular file whose size fstat reports up front. The same 839-byte file
therefore printed differently depending on how it was handed to wc:

$ wc < 839-byte-file
     18      20     839
$ wc 839-byte-file
 18  20 839 839-byte-file

GNU wc prints 18 20 839 in both cases; it only falls back to the minimum
width when the size genuinely is not known in advance.

This PR takes the size of stdin from its metadata when stdin is a regular file
and feeds it into the existing width computation, so:

• wc < file matches wc file;
• a - argument contributes the size of stdin to the total, the way a named
  file does (wc - other-file < file widens to fit size(file) + size(other));
• pipes, terminals and character devices are unchanged — cat file | wc and
  wc < /dev/null still use the minimum width of 7;
• a regular file of size zero still yields width 1, matching how a zero-byte
  named file behaves.

The digit-width calculation that was inline in compute_number_width is pulled
out into a small width_of helper so both the stdin-only and the mixed-input
paths share it.

Testing: new test_stdin_size_dictates_width in tests/by-util/test_wc.rs
covers the three cases (regular file on stdin, the same content piped in, and
- mixed with a named file). It fails on the current main and passes with
this change. The full wc suite is green (58 passed), cargo fmt --check and
cargo clippy -p uu_wc --all-targets are clean.

Behaviour was established by observing GNU wc's output on a Debian system —
no GNU source was consulted.

[Socialpranker]

Side-by-side check against GNU wc (Debian, coreutils 9.x) with the patched binary built from this branch — all eight cases that differed before now match byte for byte:

OK   wc < p839            → [ 18  20 839]
OK   wc - < p839          → [ 18  20 839 -]
OK   wc - p839 < p839     → [  18   20  839 -|  18   20  839 p839|  36   40 1678 total]
OK   wc < /dev/null       → [      0       0       0]
OK   wc < big             → [ 156 1129 8011]
OK   wc -l < p839         → [18]
OK   wc < /proc/cpuinfo   → [45 310 1715]
OK   cat p839 | wc        → [     18      20     839]

/proc/cpuinfo is the interesting one: it is a regular file of size 0, so the width stays 1 even though 1715 bytes are read — same as GNU.

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/date/date-locale-hour (fails in this run but passes in the 'main' branch)
Skip an intermittent issue tests/tail/retry (fails in this run but passes in the 'main' branch)
Skipping an intermittent issue tests/tail/follow-name (passes in this run but fails in the 'main' branch)
Congrats! The gnu test tests/tail/flush-initial is no longer failing!

[sylvestre]

A bunch of jobs are failing

[sylvestre]

Is there a bug somewhere about this issue?