Part of WT-18574, on the block-cache-depr feature branch. Test-only: src/ is untouched. The legacy cache is soft-disabled (WT-15663), so random runs never exercise it — this only stops format from requesting it.

Work

• Remove the block_cache* definitions from test/format/config.sh and regenerate format_config_def.c / the V_GLOBAL_* macros — no hand edits.
• Remove the format_config.c consumers: BC_CHECK("block_cache", ...), the cache-size adjustment, the config_explicit short-circuit, the config_off turn-off entry.
• Remove the connection-string emission in wts.c.
• Strip block_cache* lines from CONFIG.disagg, CONFIG.endian and the 53 failure configs — some indent their keys (e.g. CONFIG.WT-10875), so match optional leading whitespace. The five block_cache=1 configs already ran soft-disabled, so removing the key doesn't change what they reproduce — say so in the PR.
• Keep the compatibility script's block_cache=0 line while any tested older release's format binary has the option; newer format warns and ignores unknown keys.

Guardrails

• config_single ignores unknown bare and dotted keys with a warning — historical CONFIG files replay with no compatibility-table entry.
• config_off/config_explicit use fatal=true lookups: any leftover block_cache* call aborts format at startup — remove every one in this PR.
• WT-18657 adds disagg.victim_cache* in the same files but different entries — preserve its options; no conflicts expected, just regenerate from the merged config.sh.
• Update any Evergreen task passing these options in the same PR.

Acceptance

• No V_GLOBAL_BLOCK_CACHE* macros or block_cache* consumers remain; random runs never enable the legacy cache.
• Historical CONFIG files (bare and dotted keys, including one block_cache=1 config) replay with only a warning.
• All compatibility release pairs pass; dependent Evergreen task arguments updated in the same PR.
• No src/ changes.

Dependencies

None. Parallel with T2; blocks Phase S.