spec-guards: stale seals for depin/prove and sandbox/session_timeout after #6954 and #6820 #7356

Problem

spec-guards (tools/check_seal_currency.py) fails on every PR with two stale seals that recent master merges left behind. Master's own spec-guards runs keep getting cancelled, so master does not show it.

All five hashes (spec + four gen) differ in both, measured with t27c seal --verify on the t27c Railway lab at master 223c28c.

Honesty of a reseal

• session_timeout: tests now run and pass 6/6 (they were BLOCKED before 28 modules are imported by 30 specs and do not exist (found by the oracle) #6820 removed the missing timestamp import). Safe to reseal.
• prove: tests are BLOCKED (use tri::crypto::sha256::TriSha256 does not resolve), as in the old seal. But the tests Implement part 2 of 2 of the empty bodies in specs/depin/prove.t27 (3 functions) #6954 added are wrong on their own: they use return false / return valid in a test body (Zig: expected anyerror!void, found bool), and test_pack_gf16_matrix_identity expects 0x11 at packed[j*8] while row j's diagonal nibble sits at byte j*8 + j/2. Measured in a scratch spec on the lab: the original pack tests do not compile; corrected asserts pass 3/3; a mutant without the << 4 fails 2 of 3.

Fix

1. Rewrite the Implement part 2 of 2 of the empty bodies in specs/depin/prove.t27 (3 functions) #6954 tests in prove.t27 as asserts with the correct identity expectation, plus a nibble-order test.
2. Reseal both specs on the lab; prove stays recorded as blocked (not passed) until TriSha256 resolves.

Left open

The Merkle tests in prove.t27 cannot run until tri::crypto::sha256::TriSha256 resolves; merkle_root also relies on slice ++ and [] literals that have not been compiled by any test yet.