Option 'reduceValues' to give users control over the object constructed for eah row

Nobody has claimed this yet.

## Feature Proposal

Provide a new option to the options object that allows users to choose how to modify the output object based on each cell in the row. Documentation provided below:

### reduceValues

Type: `Function`

A function that can be used to modify the object that is emitted by the stream, similar to a [reducer function](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/reduce). The return value will replace the existing object used accumulating columns in the row. If `null` is returned, the rest of the row is skipped (similar to mapHeaders).

```javascript
csv({
reduceValues: ({ memo, header, index, value }) => memo[header] === '' ? (memo[header] = value, memo) : memo
})
```

### Parameters

**memo** _Object or any_ The current object representing the values in a row
**header** _String_ The current column header.
**index** _Number_ The current column index.
**value** _String or any_ The current column value (or content).

If both `mapValues` and `reduceValues` functions are provided, `mapValues` is run first, and the output is provided in the **value** parameter of the `reduceValues` function.

## Feature Use Case

This feature gives the developer more control over constructing the object that is passed through the stream, or choosing to skip a row based on the values encountered for a row. Some specific example use case:

- If multiple columns have duplicate-named headers, instead of automatically taking the last one, a user could choose to take the one that does not have an empty value, or take the max of the values.
- Skip records (rows) if the first column is blank
- If headers are named like a JS object path, eg. `first[0].inner`, user can handle logic of constructing nested objects
- Rehydrating a row with a particular type of object, ie. on `index === 0`, instantiate a `new RowObject()`

The goal of this proposal is to allow developers more freedom over the deserialization without increasing maintenance burden for the maintainers.

** Why not just make the user transform the object downstream from the CSV parser. After all, this is only a CSV parser!!! **

Moving the object reducing logic upstream can help maintain the fast speed this library is known for, as well as allow users to build their own workaround for issues like #150. If a user has more control over skipping a row, or can stop deserializing the rest of the row, there are performance advantages, especially in very wide data sets or ones with lots of rows to be skipped. In addition, some behaviors, like when there are duplicate header names, cannot be addressed downstream.

Start by tracing the existing mapValues and mapHeaders option handling and the row-construction path. Compare those behaviors with the proposed reduceValues parameters, return values, ordering, and duplicate-header use cases. Done means the documented option behavior is implemented and its interaction with mapValues and row skipping is verified.

Written by the indexing model from the issue text.