Racy context cancellation

[azavorotnii]

Current cancellation approach have couple racy cases. I propose PR to fix them:
#744

Issues details.

1. Context cancellation is performed in separate go-routine which can live after query execution completes (sqlite3.go:1852):

	if ctxdone := ctx.Done(); ctxdone != nil {
		go func(db *C.sqlite3) {
			select {
			case <-ctxdone:
				select {
				case <-rows.done:
				default:
// <-- here query can complete and next query can start execution 
					C.sqlite3_interrupt(db)
					rows.Close() // < -- see case 2
				}
			case <-rows.done:
			}
		}(s.c.db)

This is rare case but it happen in our production environment sometimes. It causes next query executed on same connection to get "interrupted" error when its context was not cancelled.

See TestQueryRowContextCancelParallel in PR, it should reproduces issue almost always.

---

2. Closing rows after interruption can cause QueryRowContext() to return sql.ErrNoRows, as (sql.go:3019):

func (r *Row) Scan(dest ...interface{}) error {
...
	if !r.rows.Next() {
		if err := r.rows.Err(); err != nil {
			return err
		}
		return ErrNoRows
	}

Because we close rows ourselves, rows.Next() returns io.EOF which is used by "sql" package both for rows.Next() and rows.Err() results.

See TestQueryRowContextCancel in PR.

[rittneje]

Note that sqlite3_interrupt was intended for a GUI and does not have the ability to actually cancel a particular query (see #648). One way this can cause a problem is if the context expires before the goroutine running execSync even starts. In this situation, sqlite3_interrupt will be a no-op, and sqlite3_step will run to completion, no matter how long it takes.

In short, there are still races with sqlite3_interrupt even with your fix, so please be careful how you use context.Context with this library. But your fix certainly seems reasonable.

[azavorotnii]

One way this can cause a problem is if the context expires before the goroutine running execSync even starts

Good point, will put comment about it. (DONE).

[azavorotnii]

@mattn, @gjrtimmer do you still maintain this repo?