feat: allow different json library imports

[EvanR-Dev]

Allow users to import orjson in their requirements.txt - the library will use that over the standard json import if installed. Here is a sample measure of performance differences. The following code calculates the average time for how long it takes to deserialize and serialize a Cosmos document JSON (~5 KB), over 10,000 runs

Note: measured in microseconds, not milliseconds

@app.route(route="items", auth_level=func.AuthLevel.ANONYMOUS)
@app.cosmos_db_input(arg_name="inputDocument", 
                     database_name="test",
                     container_name="items",
                     id="test-json",
                     connection="pythonworker37cdb_DOCUMENTDB")
def test_function(req: func.HttpRequest, inputDocument: func.DocumentList) -> str:
    if not inputDocument:
        return "Not found"
    
    runs = 10_000
    start = time.perf_counter()
    for _ in range(runs):
        json_string = func.Document.to_json(inputDocument[0])
        _ = func.Document.from_json(json_string)

    end = time.perf_counter()
    avg_ms = ((end - start) / runs) * 1_000_000

    logging.info(f"Time taken: {avg_ms:.2f}ms")

    return "Item fetched successfully"

import json

import orjson

(Not being added, but just for comparison):
import ujson

import simplejson

[gavin-aguiar]

Update the description and keep only ORJSON

[EvanR-Dev]

Orjson is especially effective when the JSON payload increases in size. The improvements range from 15% (~1 KB payload) to even above 100% as the payload size increases

In these scenarios we have an EventHub trigger with an EventHub output binding that receives HTTP requests and sends it to an EventHub after deserializing the payload

Orjson:

Std json:

(some were not able to respond back in time as well)

If we purposely deserialize and serialize 10,000 times using the Cosmos library and measure the avg time (again)

Orjson:

Std json: