fix(sql): fill []*T in Select and skip unexported struct fields

[MrBeldum]

Fixes #4426.

1. []*T destinations came back as N nil pointers. selectSlice only routed rows through rowsToStruct when the element kind was Struct. For *T it called rows.Scan with a **T, dropped the error, and appended a nil pointer per row. Now a *struct element allocates a new T per row, fills it via rowsToStruct, and appends the pointer.

2. Unexported fields panicked. rowsToStruct called v.Field(i).Addr().Interface() on every field matching a column, which panics for unexported fields (reflect.Value.Interface: cannot return value obtained from unexported field or method). Unexported fields and fields tagged db:"-" are now skipped, the same as encoding/json and sqlx. Their columns are scanned into a throwaway value.

3. Scan errors are logged (error scanning row: ...) instead of being discarded with _ =, in both the scalar-slice and struct paths.

Tests in pkg/gofr/datasource/sql/db_test.go:

• TestDB_SelectMultiRowIntoSliceOfPointers: []*user gets filled pointers, tag mapping included.
• TestDB_SelectSkipsUnexportedAndIgnoredFields: no panic; id (unexported) and Secret (db:"-") stay zero.
• TestDB_SelectLogsScanErrors: a non-numeric value scanned into int is logged for both slice and struct destinations.

Without the change, the first two tests fail. go test, go vet, and staticcheck pass for ./pkg/gofr/datasource/sql/.

[NitinKumar004]

Thanks for picking up #4426, @MrBeldum. I checked it at b92f22b6 with sqlmock and against a real postgres:16, using the same app built from development and from this branch.

What I verified

• []*T into a row struct now works. On Postgres, []*user goes from [nil nil] on development to filled structs, with tag mapping.
• Unexported fields no longer panic, and scan errors are now logged instead of dropped. On Postgres a NULL name scanned into a string field now logs error scanning row: sql: Scan error on column index 1 ..., where development was silent.
• gofmt, go vet and go test -race -count=3 ./pkg/gofr/datasource/sql/ are clean. golangci-lint v2.12.2 reports 0 new issues.
• Mutations: removing the []*T case, the IsExported skip, or either of the two new log calls fails the new tests.

Blocking: []*time.Time, []*sql.NullString and other []*Scanner destinations now come back as zero values

The new []*T case catches every pointer to a struct. But time.Time and sql.Scanner types such as sql.NullString, sql.NullTime or decimal.Decimal are structs that scan a single column themselves. On development they went through rows.Scan, which handles **time.Time and **sql.NullString fine. Now they go through rowsToStruct instead. That finds no exported field matching the column, scans the column into the throwaway value, and appends a zero value, without logging anything.

Real postgres:16, table users(id int, name text, created_at timestamptz) with rows (1,'alice','2026-10-06') and (2,NULL,'2026-10-07'):

destination	development	this PR
[]*time.Time (SELECT created_at)	2026-10-06, 2026-10-07	0001-01-01, 0001-01-01
[]*sql.NullString (SELECT name)	"alice"/valid, nil	""/invalid, ""/invalid
[]*int64	7 8	7 8

The same shows up with sqlmock. A small helper fixes it, and it also fixes []time.Time / []sql.NullString, which were already zero-valued on development because they hit the Struct case:

// isRowStruct reports whether Select maps a row's columns onto t's fields. Struct types that scan a
// single column themselves, time.Time and sql.Scanner implementations such as sql.NullString, are
// scanned directly instead.
func isRowStruct(t reflect.Type) bool {
	return t.Kind() == reflect.Struct && t != reflect.TypeFor[time.Time]() &&
		!reflect.PointerTo(t).Implements(reflect.TypeFor[sql.Scanner]())
}

		case isRowStruct(elemType):
			...
		case elemType.Kind() == reflect.Pointer && isRowStruct(elemType.Elem()):
			...

With that applied on top of this branch, the whole sql package passes. The same probe then gives:

[]*time.Time      -> [2026-10-06T12:00:00Z 2026-10-06T12:00:00Z]
[]*sql.NullString -> [alice/valid nil]
[]time.Time       -> [2026-10-06T12:00:00Z 2026-10-06T12:00:00Z]
[]sql.NullString  -> [{alice true} { false}]
[]*int64          -> [7 8]

Could you add those destinations as rows in a test next to TestDB_SelectMultiRowIntoSliceOfPointers? Then the regression stays pinned.

Smaller notes

• nit: the tag == "-" check doesn't change behaviour today. Without it, a db:"-" field maps to a column literally named -, so it is skipped anyway, and removing the check leaves every test green. The intent is clearer with it, so I'd keep it. But the Secret row in TestDB_SelectSkipsUnexportedAndIgnoredFields can only catch a regression if the column it uses is named -.
• Worth a line in the PR description: scan errors now log at ERROR once per row. Code that maps a nullable column into a plain string will start logging where it was silent before. I think that's the right call, since the row was being silently truncated at the first failing column, but it's a visible change.

Requesting changes for the []*time.Time / []*Scanner regression. The rest looks good.