fix(codegen): no spurious lib/pq import for MySQL with sqlc.narg + sqlc.slice (#3783)

[luongs3]

Problem

A MySQL query that references the same named parameter as both a regular argument (e.g. sqlc.narg('x') IS NULL) and as sqlc.slice('x') produces Go code that imports github.com/lib/pq (a Postgres library) even though the import is unused — the runtime path expands the slice in-place via the /*SLICE:...*/? placeholder and never calls pq.Array.

Closes #3783.

Minimal repro (from the issue)

-- name: CausesPgToBeImported :many
SELECT id AS author_id, name AS author_name, bio AS author_bio
FROM authors
WHERE (sqlc.narg('author_ids') IS NULL OR id IN (sqlc.slice('author_ids')));

Generated MySQL output on main contains:

import (
    "context"
    "database/sql"
    "strings"

    "github.com/lib/pq"  // unused
)

Root cause

The two references to author_ids produce two goColumn entries with the same name and type; only the slice variant carries the IsSqlcSlice flag. The non-slice duplicate field still satisfies the []T && !sqlc.slice condition in sliceScan() inside internal/codegen/golang/imports.go, which adds lib/pq to the import set. The struct itself is rendered through UniqueFields so only one field is emitted, but the import-bookkeeping pass sees both fields.

Fix

In sliceScan(), first collect the names of all fields carrying IsSqlcSlice, then skip any field whose name matches one of those when deciding whether to require lib/pq. The duplicate is already handled by the slice-expansion code path and does not need pq.Array.

Tests

• New endtoend fixture internal/endtoend/testdata/mysql_slice_narg_no_pq_3783/mysql mirrors the issue's repro; golden files confirm no lib/pq import in the generated code.
• Full internal/endtoend TestReplay suite passes.
• go test ./internal/codegen/... ./internal/compiler/... ./internal/sql/... clean.
• gofmt -l clean, go vet clean on touched files.

What does NOT change

• Postgres / pgx output is unaffected — the field-name filter only suppresses lib/pq when a sqlc.slice with the same name exists in the same query, and pgx never went through the lib/pq branch anyway.
• MySQL queries that use sqlc.slice alone, a plain []T argument, or a sqlc.narg without a matching sqlc.slice are unchanged.