Socket sources: Improvements for internal buffers

[AmmarAbouZor]

This PR includes:

• Increase the capacity of the internal buffer in socket byte sources to 1MB instead of the default value 8KB.
• Add Checks for socket byte source to avoid filling the internal buffer if the server is sending the data with high frequency because the producer loop will call load on each iteration while some parsers will parse one item only on each iteration.
• Call flush on consume when there is not enough writable memory.
• Another check if the buffer is filled and return an error if we can't add all the incoming bytes into the buffer.

[kruss]

LGTM!

Note: Probably to avoid duplication of the logic using the buffer for UDP/TCP cases would be an improvement.

[AmmarAbouZor]

I've created two tests to demonstrate two issues:

• TCP test where the DeqBuffer is panicking
• UDP test prints how many bytes won't be parsed with parsers not consuming all the available bytes

[AmmarAbouZor]

• TCP test is fixed to not reproduce the panic. However, this shows us that the safe functions of DeqBuffer could produce an undefined behavior if miss used.
• UDP test is working now, but it still printing how many bytes won't be parsed
• The ultimate fix for the cumulated bytes is to change parsers implementation to parse as possible from the provided slice of bytes and to not parse the first item only. I think This change is out of scope of this PR because it would introduce bigger changes + performance lose with the current implementation of the producer loop

[marcmo]

LGTM
just one comment