fix: make vet accept %w in Errorf, document Wrap(nil) contract

[ernado]

What

Two related hardening changes around the library's core contracts:

1. Document and pin the Wrap(nil) contract. Wrap/Wrapf always return a non-nil error, even for a nil input — matching fmt.Errorf with a nil %w operand, and unlike pkg/errors. This was deliberate but undocumented. Now stated in godoc and README (with the guarded if err != nil idiom), and pinned by TestWrapNil so it cannot change silently. Changing it was considered and rejected: it would silently alter control flow in every downstream module, and cmd/gowrapper's Errorf→Wrap rewrite is only semantics-preserving because both sides agree on nil.

2. Fix the vet false positive on Errorf with %w. Errorf genuinely supports %w by falling back to fmt.Errorf (asserted by TestFormatError), but newer toolchains failed go test ./... with errors.Errorf does not support error-wrapping directive %w. Vet classifies printf wrappers by the calls that forward (format, a...); the fmt.Sprintf forward on the traced path downgraded Errorf from errorf-like to printf-like. Moving the Sprintf call behind a non-variadic helper leaves fmt.Errorf as the only forward, so vet now classifies Errorf correctly. No behavior change, and the same false positive disappears for downstream users writing errors.Errorf("...: %w", err).

Test plan

• go vet ./... — clean (previously failed on format_test.go:488)
• go test ./... — passes with vet enabled, including new TestWrapNil