t27b: the greedy blockers order, its replay and FNV-1a come from specs/tri/t27b/blockers.t27 (port slice of #6198) #7673

Port slice of #6198 (t27b written in t27), under #6063 and #5980.

cli/t27b/src/blockers.rs (719 lines on master 0fbb0a0) is mostly I/O: it spawns and times out processes (run_capture, kill_group), runs the reference path and keeps its cache file (RefRunner, CacheWriter, read_cache_tests, use_closure, dir_bytes), and parses tool output into Strings. The blocker names themselves (ExprCall(@abs), type u128) are produced in lower.rs, not here.

Its pure decision logic is three functions:

• greedy: the set-cover order of t27b corpus --blockers, that is, which construct to support next and how many whole files each step unlocks;
• replay: the cumulative count of such an order over another list of files;
• fnv64: FNV-1a 64, the hash behind every reference-cache key.

The plan:

• Write the three in specs/tri/t27b/blockers.t27. The files become the rows of a 0/1 matrix over the constructs, numbered in name order. The tie-break order is unchanged: files unlocked, then the progress score (summed in file order), then the files needing the construct, then the name. FNV-1a reads a byte slice.
• Generate gen/rust/tri/t27b/blockers.rs with t27c gen-rust and mount it with #[path] in blockers.rs. Delete the hand-written bodies. The public signatures stay, so main.rs, tests/blockers.rs and tests/refcache.rs do not change.

Oracle and checks

• The current Rust is the oracle. The orders, counts and hashes in the spec's tests were printed by the hand-written greedy, replay and fnv64 on the t27c Railway lab.
• t27c test-report with 0 vacuous passes, tri mutate spec, and a seal.
• cargo test --release -p t27b.
• A t27b lab corpus run of head that equals master's, file by file.
• The hand-glue budget (Only-t27 gate: a foreign-line budget the label cannot lift; local hook reads exceptions from origin/master #7371) holds, and the AGENTS.md t27b remainder line shrinks.