test: parallelize substrate e2e suite

[EItanya]

Summary

• mark all 20 top-level Go E2E tests parallel
• scale the CI Substrate worker pool from 1 to 4 and run up to 4 tests concurrently
• use a conservative parallelism of 2 for local E2E runs, where the default install has one worker
• remove the CLI test environment mutation that prevented parallel execution

Validation

• E2E package compile: passed
• 42/42 compatible non-Claude test executions passed across three shuffled, uncached four-worker runs with -parallel 4
• first run: 14 tests in 65.31s
• two-run stress: 28 tests in 150.28s
• shared-agent cases reached about 40s when all four workers were occupied, but did not time out
• mixed full-suite run exercised concurrent actor lifecycle work; Claude cases were blocked by the reused cluster controller and an existing CRD fixture mismatch
• git diff --check: passed

Follow-up

This reduces suite wall-clock time but does not remove the individual ~20 second latency. That is caused by the fixed 20 second ActorTemplate reconciliation poll in Substrate v0.0.22 and needs an upstream fix.