Elazar decision: authorize coder03-mars-ca + db03-mars-ca as rows in mars-commons.md, or the codex switch cannot land code on Mars

BLOCKS THE CLAUDE-CODE-TO-CODEX SWITCH ON MARS. 73 open mars rows were routed to coder03-mars-ca (51) and db03-mars-ca (22) on 2026-09-13 after Elazar directed fleet work onto codex seats (cc weekly meter 24% left, codex 99%). Both seats are on analysis/bwi-only hold because mars-commons.md section 'Agent Code-Edit Authorization' names exactly three Mars agents - coder-mars-cc, coder02-mars-cc, db-mars-cc - and states verbatim: 'There is no separate pm-mars-cc seat and no -ca twin of any Mars lane', 'A push grant is a row in THIS file and nowhere else', and 'A verbal grant from the PM is not a grant.' My routing was a PM DM, which is the named non-mechanism; coder02-mars-cc declined a gate role and cited it. The same gate caught the same error on 2026-09-11 when the PM told coder02-mars-cc to push and it correctly filed a handoff commit instead. TENSION IS BETWEEN TWO OF ELAZAR'S OWN INSTRUCTIONS, which is why no agent may resolve it: EVO-101 (his roster design) defines coder03-(project) as a real seat that both codes and audits, while mars-commons.md:57 says no -ca twin of any Mars lane exists. NOT SELF-FIXABLE BY A PM: the authorization IS the file, so a PM editing the file that says a PM's word is not a grant would be a self-grant. Needs Elazar's own edit adding coder03-mars-ca and db03-mars-ca as rows, with their push/DDL scope stated. Related: db02-mars-cc has ACCEPTED the pre-apply DDL review gate for db03-mars-ca, so the review half is ready the moment authorization lands.