[BUG] Loader incorrectly creates a new batch before reaching maxBatchSize and batch dispatch

[leonbe02]

Expected Behavior

If I try to load multiple keys (some being duplicates) before a batch is dispatched, I would expect that my batch function would trigger Math.ceil(queueSize / maxBatchSize) times after the intentional delay using batchScheduleFn. queueSize being the number of unique keys that have yet to be dispatched.

Current Behavior

The getCurrentBatch function is creating a new batch when existingBatch.cacheHits.length reaches the maxBatchSize, thus spreading out the keys across more batches than are needed.

Possible Solution

https://github.com/graphql/dataloader/blob/master/src/index.js#L92
On the line above, it is currently pushing to the cacheHits array, regardless of if the key is a duplicate of a previous load on the same batch. If this were instead a Map<key, fn>, it would prevent duplicate loads from incrementing this array's length and causing new batches to be created when they aren't needed.

Steps to Reproduce

const loader = new DataLoader(async (keys) => {
  console.log(keys);
}, {
  maxBatchSize: 3,
  batchScheduleFn: (callback) => setTimeout(callback, 100)
});

const keys = ['a', 'b', 'a', 'a', 'a', 'b', 'c'];
for (const key of keys) {
  loader.load(key);
}

/*
The code above will log 2 separate batches:
[ 'a' , 'b' ]
[ 'c' ]
Instead of the expected single batch of:
[ 'a', 'b', 'c' ]
*/

Context

We are trying to batch as many IDs as we can into a single request using DataLoader, however, since we are using Apollo GraphQL, there are times where a query might attempt to load the same ID for different fields within the schema. When this happens enough times, it reaches the maxBatchSize and causes the issue I outlined above where new batches are being created when they don't need to be.