Use draining read in standard.c++ pumpTo impl

[jasnell]

Part of ongoing incremental perf improvements. This updates the standard->to->internal pumpTo loop (consuming a JS readable stream and writing the data to a kj destination) to use the new draining read optimization.

In a simple local benchmark, we see...

| Metric         | Main      | New       | Improvement       |
| -------------- | --------- | --------- | ----------------- |
| Avg Req/Sec    | 777       | 5,026     | 6.5x (+547%)      |
| Avg Latency    | 12.34 ms  | 1.35 ms   | 9.1x lower (-89%) |
| p50 Latency    | 10 ms     | 1 ms      | 10x lower         |
| p99 Latency    | 45 ms     | 8 ms      | 5.6x lower        |
| Max Latency    | 1,771 ms  | 22 ms     | 80x lower         |
| Avg Throughput | 1.49 MB/s | 7.19 MB/s | 4.8x              |
| Total Requests | 9k        | 55k       | 6.1x              |

the number vary broadly based on chunk size, write pattern, etc but in general we should see significant improvements across the board.

The specific scenario that is improved are things like return new Response(new ReadableStream(...)) in a worker fetch handler. That takes the JS-backed ReadableStream and drains it into the KJ stream at the internal level.

The specific optimization here is that we no longer perform a 1-to-1 read/write loop but instead drain as much data as we can synchronously off the JS-backed ReadableStream every time we are holding the isolate loop the coalescing writes to the kj layer. It means fewer JS->kj iterations, fewer times that we have to take the isolate lock, fewer promise allocations, fewer data copies, etc.