Avoid floating point false alarms in the TCP model's missing ACK assertion

[nfrisby]

Fixes #393.

[nfrisby]

@bwbush this runs out of memory even on my 32GB machine, so I can't test my theory.

I've pushed a couple commits so you can.

• First commit skips the "triangle inequality check", which was a slow distraction for this Issue.
• Second commit rearranges the arithmetic/inequality in case it's a false alarm caused by floating point issues.
• Second commit also now prints out the offending numbers for future debugging.

[bwbush]

This executes correctly at 327ba1c far past the point at which the assertion failed in issue #393, but using the same configuration and network topology.

[nfrisby]

Great. I'll tidy it up and merge it today.

[nfrisby]

@bwbush FYI this PR also adds the --skip-triangle-inequality-check to the sim subcommand, since its time complexity is painful with large topologies and it's not particularly critical. (And apparently the real world data doesn't satisfy it.)