doc: DecodeHook's stated default is the hook the code replaced

[ump45nose]

Overview

DecodeHook's doc comment states the wrong default decode hook. One-line doc fix.

What problem does it solve?

viper.go:58-64 documents the default as:

//	 mapstructure.ComposeDecodeHookFunc(
//			mapstructure.StringToTimeDurationHookFunc(),
//			mapstructure.StringToSliceHookFunc(","),
//		)

but defaultDecoderConfig — the function that actually establishes the default, at viper.go:1003-1007 — uses Viper's own weak-slice hook, and the mapstructure one is explicitly commented out right above it:

decodeHook = mapstructure.ComposeDecodeHookFunc(
	mapstructure.StringToTimeDurationHookFunc(),
	// mapstructure.StringToSliceHookFunc(","),
	stringToWeakSliceHookFunc(","),
)

stringToWeakSliceHookFunc is described at viper.go:1026-1028: mapstructure v2.0.0's StringToSliceHookFunc checks that the target is a string slice, and this helper removes that check (viper.go:1035 only tests t.Kind() != reflect.Slice), so comma-splitting also reaches non-string slice fields, whose elements are then converted by WeaklyTypedInput. That is a real behavioural difference, not a naming nit — and this text is what shows up on pkg.go.dev for DecodeHook.

Change

Point the documented example at the hook that is actually used:

//	 mapstructure.ComposeDecodeHookFunc(
//			mapstructure.StringToTimeDurationHookFunc(),
//			stringToWeakSliceHookFunc(","),
//		)

The other three mentions of StringToSliceHookFunc in the repo are left alone because they are correct in place: viper.go:1005 is the commented-out line being replaced, viper.go:1026 is the comment explaining why it was replaced, and viper_test.go:900,938 pass the strict hook explicitly to test the override path.

Tests

Comment-only, so no test change is needed and no behaviour moves. I did not run go build / go test — the command layer in this session refused all build and test invocations, so I am stating that rather than implying CI ran locally. The change is inside a // doc comment above func DecodeHook, which cannot affect compilation; gofmt-relevant whitespace was preserved exactly (the tab-indented //\t block is untouched apart from the identifier).

Related issues

None. Checked the open PR list before starting: #2143 ("Export DefaultDecodeHook for composition") touches this same default in a different way — it extracts defaultDecoderConfig's chain into an exported DefaultDecodeHook() and does not edit lines 58-64 — so the two do not overlap textually, and this correction is worth landing either way. If #2143 merges first, the doc here should probably be reworded to name DefaultDecodeHook() instead of the inlined chain.

Note: this PR was prepared and submitted by an autonomous agent on behalf of ump45nose (Viper has no AI-disclosure rule, but the authorship seems worth stating plainly for a doc change).