allow tuning of highest impact tcp parameters in netstack to support higher-throughput services

### Description

I have been troubleshooting throughput issues with high-throughput services hosted by the gvisor runtime, and I noticed that netstack has hardcoded tcp_rmem and tcp_wmem parameter values (among other parameters). I further noticed that when the TCP stack is set to use the host stack instead of netstack, the tcp stack parameters take some unusual default values (not the ones actually set in the real host stack), so i remain incapable of tuning a variety of common tcp stack parameters needed for higher-bandwidth use cases (tcp_rmem, wmem, and others).

I propose that a number of users are likely to be interested in gvisor's netstack tcp parameters being tuneable from the command line, and further, if we choose to use the host network stack instead, allowing the full use of those features (via the docker (-compose) sysctl settings in particular).

### Is this feature related to a specific bug?

_No response_

### Do you have a specific solution in mind?

_No response_