Strong-type data points, avoid magic strings

It is great that this project is being converted to TypeScript. I wanted to suggest to use generics for strong-typing data points in time series, and avoid magic strings as much as possible. I created this commit to show how the interfaces could be updated to do that (it does not include the implementation):

Commit: [Add generic type for data points](https://github.com/omidkrad/pond/commit/4e4818dfbc5233d08a040e821b0392babef22324)

With that, as an example, `avg` function could be updated to support this signature:
```typescript
avg(key: (data: TData) => number, filter?: any): number;
```

and be used like below:

```typescript
import { timeSeries } from "pondjs";

interface SensorDataPoint {
time: number;
sensor: number;
status: string;
}

const series = timeSeries({
name: "sensor_data",
columns: ["time", "sensor", "status"], // checked, non-existent keys like "status1" will give error
points: [
[1400425947000, 3, "ok"],
[1400425948000, 8, "high"],
[1400425949000, 2, "low"],
[1400425950000, 3, "ok"],
]
});

// by field
series.avg(d => d.sensor); // 4

// calculation
series.avg(d => d.sensor + d.sensor2 * 100);

// field path
series.avg(d => d.sensor.diningRoom);
```

I believe function signatures that receive magic strings should be deprecated in favor of this new signature.