Report the flag name as typed in deprecation warnings

[hdimer]

Fixes #279.

With an alias from SetNormalizeFunc, using the alias for a deprecated flag printed the canonical name rather than the one typed:

--dir=test  ->  Flag --src has been deprecated, ...   # before
--dir=test  ->  Flag --dir has been deprecated, ...   # after

Parse's callback passed flag.Name into Set, so the typed name was gone before the warning was built. This threads the typed name through the internal parseFunc so the warning reports it. Long flags report the typed alias; shorthands keep the canonical long name (no alias to report). Public Set/ParseAll signatures are unchanged; calling Set directly with an alias now echoes that alias.

Added TestDeprecatedFlagUsageAlias.

Used AI assistance on this; I reviewed and tested it.

[CLAassistant]

All committers have signed the CLA.

[hdimer]

Flagging that the red Lint here isn't this change.

golangci-lint v2.7.2 can't read Go 1.27's export data (export data version 4 is greater than maximum supported version 2), so every finding is a typecheck cascade off unresolvable stdlib imports — including errors.go:44:1: missing return for a function whose last statement is panic(...), which is a valid terminating statement. A genuine compile error there would have taken the test jobs with it; all six pass, 1.12 through stable.

Not asking anyone to fix the linter — just noting the PR isn't broken. Happy to rebase if a re-run would pick up a newer golangci-lint.

[tomasaschan]

Ah, thanks for flagging. I suppose we need to bump golangci-lint now that 1.27 is out, which rolls stable/oldstable build jobs one up.

PR welcome, otherwise I'll get to it when I get to it 😅

[hdimer]

Opened #508 for the golangci-lint bump.

v2.13.2 is built with go1.27.0, which is what fixes the export-data read. One wrinkle: the bump also switches on govet's new inline analyzer, which wants reflect.Pointer over reflect.Ptr in three places — that needs Go 1.18 and go.mod still says 1.12, so I disabled the analyzer with a comment rather than touching the call sites.

Kept it to the version bump so it doesn't collide with #451.