benchmark: use plain text proxy.

[pvbouwel]

In order to not have misleading benchmark outcomes as detailed in #21 (comment) we will benchmark with a plaintext proxy from now on. This will make the process slower (quite a bit for PutObject as it wouldn't need the extra pass on the body) but the results will be more intuitive because the way the S3 SDK client sends the data will be similar.

[pvbouwel]

As shown in the benchmark report: https://github.com/VITObelgium/fakes3pp/actions/runs/13036429752/artifacts/2505676500

The HTTP proxy endures a penalty since the client calculates a payload hash:

FakeS3Proxy/S3ProxyBeforeFakeS3Backend-createRandomObject256MB-4                   939.9m ± 1%   1300.1m ± 1%  +38.32% (p=0.000 n=10)

As a result it is close (15 ms slower) to the benchmark of just the fake testing backend:

FakeS3Proxy/FakeS3Backend-createRandomObject256MB-4                                 1.279 ± 1%     1.285 ± 1%        ~ (p=0.063 n=10)

list bucket & getObject actions are in millisecond range and very close together. The proxy case for getObject reports faster than without but with the +- 1 % they overlap and the case without the proxy (which is exact the same run also shows a 5.08% increase so this observation can be attributed to benchmark variability)

list bucket performance  2.477m ± 1% (noproxy) <>2.849m ± 1% (proxy) show 
getObject performance 62.05m ± 1%  (noproxy) <> 61.06m ± 1% (proxy)

Overall I prefer this benchmark approach as the results are easier to interprete and don't require knowledge on difference in processing by the client side SDK depending on the endpoint uses TLS or not.