Reduce memory load of redistributable download

[atifaziz]

The use of HttpClient.GetAsync to download the Python redistributable in RedistributableLocator causes
the entire response content to be buffered to memory before being written to disk. Just stepping over the following line in the debugger:

CSnakes/src/CSnakes.Runtime/Locators/RedistributableLocator.cs

Line 166 in de05da4

using HttpResponseMessage response = await client.GetAsync(fileUrl);

shows the memory spike (over 2,600 objects and tens of MBs allocated):

If one examines the heap for byte arrays, we find a 41.6MB allocation done by HttpContent.LimitMemoryStream

The exact allocation size of 416,204,438 corresponds to the HTTP content length:

The PR fixes the problem by using HttpClient.GetStreamAsync, which checks for a success response code before returning a stream over the content (instead of entirely buffering it). The image below shows the memory usage drops dramatically, allocating just 284.4 KB after stepping over the HttpClient.GetStreamAsync call (ID 2) and something similar until the end of the DownloadFileToTempDirectoryAsync method (ID 3):