fix(ci): pin consensus-spec-tests fixture to v1.6.0-alpha.6

[pauldelucia]

Summary

`ci/circleci: test` has been silently failing on master for ~7 months
because the Makefile downloads `mainnet.tar.gz` from "the latest
`consensus-spec-tests` release" (`head -n 1` of the releases API),
unpinned. `testing/ef-tests/tests/build_proofs.rs` hardcodes expected
SSZ proofs over `/ssz_random/` fixtures, which are regenerated with
fresh randomness on every release — so the unpinned download rots
the hardcoded proofs the moment a new release ships.

Timeline

Date	Event
2025-09-23 06:48 UTC	PR #1896 merged; `build_proofs.rs` last touched. CI green (CI 40491)
2025-09-23 12:33 UTC	PR #1897 merged. CI still green (CI 40514)
2025-09-24 14:11 UTC	`consensus-spec-tests v1.6.0-beta.0` released — new randomness in `bellatrix/ssz_static/.../ssz_random/case_0/value.yaml`
2026-04-28 onward	Every PR's CI fetches the post-Sep-24 fixtures; all four bellatrix proof tests fail with "left != right" Merkle path mismatches (e.g. PR #1899)

The 4 failing tests are all bellatrix random-case fixtures
(`beacon_block_body_root_proof`, `historical_batch_block_root_proof`,
`block_body_execution_payload_proof`,
`execution_payload_block_hash_proof`); the 3 passing tests use
fixtures the new release didn't perturb.

Fix

Introduce `EF_TESTS_VERSION` and construct the download URL directly
instead of scraping the GitHub releases API:

```makefile
EF_TESTS_VERSION = v1.6.0-alpha.6
EF_TESTS_URL = https://github.com/ethereum/consensus-spec-tests/releases/download/\$(EF_TESTS_VERSION)/\$(EF_TESTS_TARGET)
```

`v1.6.0-alpha.6` was the release current on 2025-09-23 (when
`build_proofs.rs` was last updated and CI was green), so this
restores the exact fixture set the hardcoded proofs were written
against. `ALL_RELEASES_URL` and the curl-grep-head pipeline are
removed since they no longer do anything.

Test plan

Verified locally — with this pin, all 7 `ef-tests` pass:

```
$ rm -rf testing/ef-tests/mainnet
$ make ef-tests
running 7 tests
test execution_payload_block_hash_proof ... ok
test beacon_block_body_root_proof ... ok
test block_body_execution_payload_proof ... ok
test historical_batch_block_root_proof ... ok
test test_historical_summaries_with_proof ... ok
test beacon_state_historical_summaries_proof_deneb ... ok
test beacon_state_historical_summaries_proof_electra ... ok

test result: ok. 7 passed; 0 failed
```

Notes

This is the follow-up to #1899 promised in that PR's
comment.
The two are independent — this one unblocks the CI on every open and
future PR; #1899 is a separate `vergen` build-deps pin needed by
downstream consumers of the published `ethportal-api 0.12` crate.

When `consensus-spec-tests` next ships a release with new randomness,
the bump should be deliberate: refresh the expected-proof arrays in
`build_proofs.rs` against the new fixtures, then bump the version
constant in the same commit. Worth considering a CI job that warns
when a new `consensus-spec-tests` release is out, but that's beyond
this PR's scope.

[pauldelucia]

Closing — trin is unmaintained per the README banner; master CI isn't being run and won't be. The diagnosis (unpinned consensus-spec-tests download in the Makefile) and the proposed pin to v1.6.0-alpha.6 (the release current when build_proofs.rs was last updated) are still correct for any fork. Cross-linked from #1899.