build: make all 76 utilities compile against current V

[metif12]

This is a prerequisite for everything else: main currently does not build, so every
CI job has been red since 2025-12-21 (last green run before that was the
Timeout implementation merge).

$ v run build.vsh
failed    cmd: v -Wimpure-v -o bin/expand ./expand
expand/expand.v:11:17: error: if condition must be `bool`, not `int`
...
$ make testfmt
.../src/shuf/shuf.v is not vfmt'ed
Encountered a total of: 14 formatting errors.

9 of the 76 utilities failed to compile, and v fmt -verify . rejected 14 files.
With this change all 76 build and make testfmt passes.

Compile fixes

Cause	Utilities
v.mathutil removed; min/max moved to math	basenc, cut, tail
for ; cond; post {} / for i := n; i; i-- {} now need a bool condition	cksum, expand
time.sleep takes a time.Duration	sleep
const cannot hold a function result	printenv
?(...) returns yield no err, but the callers read one	cp, mv, head
C.statvfs already declared privately by vlib/builtin/cfns.c.v	stat
struct utmpx declared twice with different layouts	users
$embed_file no longer resolves inside a function body	stat

Each is a one-line-or-so change with no behaviour change. Two deserve comment:

sleep previously wrote time.sleep(seconds * time.second). The conversion is
now explicit and saturates at max_duration, so sleep inf keeps waiting
instead of overflowing to roughly zero.

cp/mv/head build their settings with ?, but every failure path in those
functions calls the @[noreturn] common.exit_with_error_message. They can
never return an error, so the callers' or { ... err.msg() } was dead code that
could not compile. They now return plain tuples.

stat

stat needed more than a compile fix.

It hung the compiler. stat_to_vlib.v (now modes.v) imported
v.scanner and v.pref to decode --printf escapes. Building src/stat
pulled in the whole V compiler: v build never finished, and it took the
machine down with it. Measured directly:

before:  timeout after 150s, no output
after:   compiles in 2s

The escapes are now decoded by a small purpose-built routine, which also picks
up the \NNN and \xHH forms GNU supports and the old version silently
dropped.

It segfaulted on every call. statx() wrote the kernel's 224-byte
struct statx through a voidptr into a V struct of 160 bytes, so the write
ran past the end of the allocation. struct statvfs has the same problem; its
V mirror is now the full ABI including the trailing spare. The C structs are
declared and used for real now.

<bits/statx.h> is glibc-internal, so it is missing on musl. This is
issue #145; the kernel uapi <linux/stat.h> is used instead.

Spot-checked against GNU coreutils 9.4:

$ stat -c '%n %s %F' /tmp/sc.txt      $ ./bin/stat -c '%n %s %F' /tmp/sc.txt
/tmp/sc.txt 6 regular file            /tmp/sc.txt 6 regular file

$ stat --printf 'A\tB\x41C\101D\n' …  identical
$ stat -c '%N' /tmp/sc_link           identical

wc

read_chunk returned ?FileChunk while its caller inspected err, which does
not exist for an option. os.File.read also signals end of file with an
os.Eof error, and a zero-length read indexed buffer[-1], so wc on an
empty file would have panicked. It now returns !FileChunk and separates
exhaustion from a real read error. Also: \r was matched twice in the same
match, and byte(bool) is no longer a valid conversion.

Output now matches GNU for empty files, stdin, every count flag, multi-file
column alignment, and the missing-file error.

Formatting

v fmt -w . over the 16 files v fmt -verify rejected.

Known remaining failures

make test still fails, in 10 files. These are pre-existing logic bugs, not
build regressions, and each is independent:

Test file	Problem
groups_nix_test.v, id_nix_test.v	group lookup returns garbage ids and repeats root four times
numfmt_test.v	2K renders 2000.0; --to=none gives 2e+06
truncate_test.v	24G overflows to -1769803776 (32-bit size)
shuf_test.v	issue #189, --random-source does not reproduce GNU's permutation
stat_nix_test.v	one format line differs by 6 bytes
fmt_test.v, tail_test.v, string_to_i64_test.v, sum_test.v	not yet diagnosed

I would rather land the build fix on its own and take these as separate PRs, so
each is reviewable. Happy to keep going in this branch instead if you prefer.

Testing

• v run build.vsh — all 76 utilities build, 0 failures
• make testfmt — passes
• make test — 37/47 test files pass (was: 9 utilities could not build at all,
  so their tests could not run at all)

Verified on V 0.5.2 (0dc6a69, upstream vlang/v master) on Ubuntu 24.04 with
GNU coreutils 9.4, which is the same V the CI checks out.