increase max offset for crdb cluster in e2e tests

[ecordell]

analyzing recent e2e flakes suggested that:

• if the gha runner is slower (that it was when the test was written), the clock skew may be enough to mark a node as bad and be omitted from the cockroach cluster (thus ruining the test)
• write performance on the gha runners seems to have dropped, so that the steps that fill crdb with data sometimes stall out

I bumped the max-offset for the cluster which should prevent crdb nodes from being marked bad even with the offsets we use to demonstrate the new enemy problem.

I also reduced the amount of data that the test writes - it was overly aggressive before just because (at the time) it was still faster than the other test suites. This should™️ prevent the test from stalling out on writing filler data.

I re-ran e2e twice on this PR after these changes and both times they converged very quickly.