shreds-e2e: shard tests across 4 runners

[packethog]

Summary of Changes

• Shard the shreds-e2e workflow across 4 parallel runners
• setup job discovers all TestE2E_* functions in the checked-out shreds repo, pins the two heaviest tests (TestE2E_MultiUserInstantAllocation ~396s, TestE2E_DeviceScale ~345s) to their own shards, and round-robin distributes the remaining 19 tests across shards 3-4. Matrix is emitted as JSON on the job's outputs.
• e2e job becomes a matrix job consuming that output, named e2e (shard N), with timeout-minutes: 15 and go test -timeout=12m -run '${{ matrix.run }}'.
• Reduces wall-clock time of the in-repo shreds e2e run from ~15 min (one monolithic job) to parity with the shreds repo's own CI.

Testing Verification

• Ran the shard-discovery bash block locally against the current malbeclabs/doublezero-shreds checkout — produced a 4-entry matrix covering all 21 discovered tests with every test appearing in exactly one shard's regex.
• Parsed the workflow YAML with a yaml loader; confirmed the e2e job's strategy.matrix.include resolves from needs.setup.outputs.matrix.
• Actual parallel shard behavior will be verified on the PR's first CI run — expect 4 jobs named e2e (shard 1) through e2e (shard 4) running concurrently.