Reject --flag=value on no-argument boolean flags

[julesmcrt]

pflag's BoolP silently accepts explicit values on no-argument boolean flags — rm --force=false target deletes the file, head --verbose=false runs — where GNU coreutils reject --flag=value before doing anything. df already worked around this with a private helper.

This promotes that helper to builtins.NoArgBool and migrates all 24 builtins to it, so the whole shell rejects --flag=value with the GNU wording (option '--force' doesn't allow an argument). Bare flags are unchanged.

It also fixes the custom pflag.Value boolean flags that the plain Bool/BoolP migration didn't cover and which silently accepted the explicit-value form:

• grep -H/-h/-l/-L (orderedBoolFlag) — accepted --with-filename=true via strconv.ParseBool
• strings -o (octalFlagVal) — ignored -o=x
• xargs -0/--null (trackedBool) — accepted --null=true via strconv.ParseBool
• head -q/--quiet/-v/--verbose (boolSeqFlag) — used a literal "true" NoOptDefVal sentinel, which is forgeable via =true, so --quiet=true was silently accepted (only --quiet=false was rejected)

All four now use the unforgeable builtins.NoArgSentinel (NUL byte) and route through the same flagparser rewrite, so they produce GNU's option '--foo' doesn't allow an argument. head additionally gains the precise per-flag wording it previously lacked.

Closes #538.

Out of scope (follow-ups): du's seqBool still intentionally accepts --flag=false. (pwd/cd already use a NUL sentinel and reject --flag=value correctly, with a generic option doesn't allow an argument message.)

🤖 Generated with Claude Code