add glamsterdam-devnet-7-full subdir: full suite, no repricing marker, 200M/300M, benchmarks/amsterdam

[qu0b]

Adds a v1/glamsterdam-devnet-7-full subdir to the repricing context, per the ask in #benchmarking: fill the complete EEST benchmark suite (drop -m repricing) at 200M and 300M gas, from benchmarks/amsterdam (forks/amsterdam now merges devnet-8 spec PRs; benchmarks/amsterdam is its frozen pre-devnet-8 snapshot — exact ancestor commit 178d9f9, so the state-actor sender/delegate key constants and stub addresses are unchanged).

Test types keep their plain stateful/compute names — the full suite is distinguished by subdir. The existing v1/glamsterdam-devnet-7 (repricing-marked) subdir is untouched and keeps running in parallel.

Changes

• test-source.{stateful,compute}.builder.yaml: same shape as the devnet-7 builders, with eest_ref: benchmarks/amsterdam, no marker, gas_benchmark_values: [200,300]. The force: true workaround flags aren't carried over (one-time fix for leftover partial fixtures; this subdir's fixtures dir is fresh). The compute not (test_blobhash and one_blob) exclusion is kept — the filler rejects that variant INVALID.
• global.yaml: own EEST_FIXTURES_DIR (glamsterdam-devnet-7-full), following the per-subdir convention.
• clients.yaml: byte-identical copy of devnet-7's (diff -q clean) — same instances so the two suites stay comparable; keep them in sync.
• Workflows: the new subdir added to the subdir choice in build + run. No other machinery changes: release derives everything from the build run title, and dispatchoor/generate.sh already scans all repricing/v1/* subdirs — with no runner files yet it skips this one (verified: regen is a zero-line diff).
• Build-only for now: the test-source.*.runner.yaml files land in a follow-up once the first build+release exists to point fixtures_url at. That follow-up should also set run timeouts for this subdir in generate.sh.

Coverage notes (verified against the branch)

• Dropping the marker adds the unmarked tests: stateful/bloatnet/{test_call,test_create,test_erc20,test_transient_storage,depth_benchmarks}, compute/scenario/{test_mix_operations,test_transaction_types,test_unchunkified_bytecode}, and the compute/eip7928_block_level_access_lists tests (85 of 133 test functions carry the marker today).
• test_call/test_erc20 need factory_stub/erc20_stub stub addresses the state-actor prestate doesn't provide — EEST warns and skips those (verified in address_stubs.py::parametrize_args), it does not fail the fill. Filling them later needs new state-actor entities + stub entries.
• stateful/eip7928_block_level_access_lists stays excluded (test path is tests/benchmark/stateful/bloatnet, matching the devnet-7 scope) — can widen after a first green build.
• Release tags embed test-type + EEST sha but not subdir, so devnet-7 and devnet-7-full releases of the same test type are distinguished only by sha/timestamp once glamsterdam-devnet-7: pin eest_ref to frozen benchmarks/amsterdam #22 lands; pass an explicit tag input on release dispatch if clearer naming is wanted.

Ordering

Dispatch benchmarkoor-build.yaml with subdir=v1/glamsterdam-devnet-7-full only after the state-actor datadir regen (see #benchmarking), so payloads are filled against the new state. Companion: #22 pins the existing devnet-7 subdir's eest_ref to the same frozen branch.