Use RIO_TILER_MAX_THREADS in docs

docs/mosaic.md

@@ -31,7 +31,7 @@ rio_tiler.mosaic.mosaic_reader(
     *args: Any,
     pixel_selection: Union[Type[MosaicMethodBase], MosaicMethodBase] = FirstMethod,
     chunk_size: Optional[int] = None,
-    threads: int = MAX_THREADS,
+    threads: int = RIO_TILER_MAX_THREADS,
     allowed_exceptions: Tuple = (TileOutsideBounds,),
     **kwargs,
 )
@@ -127,14 +127,14 @@ mosaic_assets = ["1.tif", "2.tif", "3.tif", "4.tif", "5.tif", "6.tif"]
 for chunks in _chunks(mosaic_assets, chunk_size):
 
     # 2nd level loop - Uses threads for process each `chunck`
-    with futures.ThreadPoolExecutor(max_workers=max_threads) as executor:
+    with futures.ThreadPoolExecutor(max_workers=RIO_TILER_MAX_THREADS) as executor:
         future_tasks = [(executor.submit(_tiler, asset), asset) for asset in chunks]
 ```
 By default the chunck_size is equal to the number or threads (or the number of assets if no threads=0)
 
 #### More on threading
 
-The number of threads used can be set in the function call with the `threads=` options. By default it will be equal to `multiprocessing.cpu_count() * 5` or to the `MAX_THREADS` environment variable.
+The number of threads used can be set in the function call with the `threads=` options. By default it will be equal to `multiprocessing.cpu_count() * 5` or to the `RIO_TILER_MAX_THREADS` environment variable.
 In some case, threading can slow down your application. You can set threads to `0` or `1` to run the tiler in a loop without using a ThreadPool (ref: [#207](https://github.com/cogeotiff/rio-tiler/issues/207)).
 
 Benchmark: