Reduce QueryContext allocations by reusing the channel

[charlievieth]

This commit reduces the number of allocations and memory usage of QueryContext by inverting the goroutine: instead of processing the request in the goroutine and having it send the result, we now process the request in the method itself and goroutine is only used to interrupt the query if the context is canceled. The advantage of this approach is that we no longer need to send anything on the channel, but instead can treat the channel as a semaphore (this reduces the amount of memory allocated by this method).

Additionally, we now reuse the channel used to communicate with the goroutine which reduces the number of allocations.

This commit also adds a test that actually exercises the sqlite3_interrupt logic since the existing tests did not. Those tests cancelled the context before scanning any of the rows and could be made to pass without ever calling sqlite3_interrupt. The below version of SQLiteRows.Next passes the previous tests:

func (rc *SQLiteRows) Next(dest []driver.Value) error {
	rc.s.mu.Lock()
	defer rc.s.mu.Unlock()
	if rc.s.closed {
		return io.EOF
	}
	if err := rc.ctx.Err(); err != nil {
		return err
	}
	return rc.nextSyncLocked(dest)
}

Benchmark results:

goos: darwin
goarch: arm64
pkg: github.com/mattn/go-sqlite3
cpu: Apple M1 Max
                                          │   old.txt   │              new.txt               │
                                          │   sec/op    │   sec/op     vs base               │
Suite/BenchmarkQueryContext/Background-10   3.994µ ± 2%   4.034µ ± 1%       ~ (p=0.289 n=10)
Suite/BenchmarkQueryContext/WithCancel-10   12.02µ ± 3%   11.56µ ± 4%  -3.87% (p=0.003 n=10)
geomean                                     6.930µ        6.829µ       -1.46%

                                          │   old.txt    │                new.txt                 │
                                          │     B/op     │     B/op      vs base                  │
Suite/BenchmarkQueryContext/Background-10     400.0 ± 0%     400.0 ± 0%        ~ (p=1.000 n=10) ¹
Suite/BenchmarkQueryContext/WithCancel-10   2.376Ki ± 0%   1.025Ki ± 0%  -56.87% (p=0.000 n=10)
geomean                                       986.6          647.9       -34.33%
¹ all samples are equal

                                          │  old.txt   │               new.txt                │
                                          │ allocs/op  │ allocs/op   vs base                  │
Suite/BenchmarkQueryContext/Background-10   12.00 ± 0%   12.00 ± 0%        ~ (p=1.000 n=10) ¹
Suite/BenchmarkQueryContext/WithCancel-10   38.00 ± 0%   28.00 ± 0%  -26.32% (p=0.000 n=10)
geomean                                     21.35        18.33       -14.16%
¹ all samples are equal