Preserve positional-only signatures in funcutils wrappers

[ZelinZhu-Richard]

The change

Fixes #510. A wrapper around def f(a, /, **kw) currently raises TypeError for the valid call f(1, a=2), because its generated signature loses /.

from boltons.funcutils import wraps

def original(a, /, **kw):
    return a, kw

@wraps(original)
def wrapped(*args, **kwargs):
    return original(*args, **kwargs)

wrapped(1, a=2)  # Expected: (1, {'a': 2}); previously raised TypeError

Pass the existing positional-only metadata through the signature formatter and keep it synchronized when arguments are removed. The formatter gains an optional trailing posonlyargs argument; invocation forwarding is unchanged. Tests cover keyword-name collisions, rejected keyword binding, defaults and annotations, mixed signatures, and injected argument removal.

Validation:

• All six new regressions fail on the original code and pass with the fix.
• Full tox suite: 678 tests/doctests passed.
• Focused funcutils suite: 48 passed; git diff --check passed.
• test_wraps_posonly_keyword_collision and test_wraps_posonly_rejects_keyword_binding cover keyword collisions and rejection of keyword binding to positional-only parameters.
• test_formatargspec_posonly_defaults_and_annotations and test_wraps_posonly_mixed_signature cover defaults, annotations, variadic arguments, and generated signatures.
• test_FunctionBuilder_remove_posonly_args and test_wraps_injected_posonly_arg cover argument removal and injection.

New positional-only test definitions use guarded exec, following the existing tests. Compatibility grammar checks passed; the full supported-interpreter matrix has not been run.

AI assistance

• Harness/tooling: Codex desktop, with separate implementation and review subagents. Codex found the bug, wrote the patch and tests, ran validation, and prepared this submission.
• Model(s): GPT-6; a more specific runtime model identifier was not exposed to the agents.