docs(sequencer-client): rewrite top-level and timing READMEs

[spalladino]

Motivation

The top-level sequencer-client/README.md was years out of date — it still referred to single-block-per-slot building and made no mention of proposer pipelining or the multi-block checkpoint model. The timing-model README still documented both pipelined and non-pipelined scheduling even though the non-pipelined mode is about to be removed. New contributors (human or AI) lacked the context they need to make changes to block building.

Approach

Rewrote the top-level README from scratch following the package's readme-writer guidelines: slots / blocks / checkpoints, proposed vs checkpointed chain, an architecture diagram, the Sequencer work loop, CheckpointProposalJob lifecycle, per-block loop pseudocode, the SequencerPublisher Multicall3 bundling and sendRequestsAt semantics, events, configuration reference, and failure modes. Trimmed src/sequencer/README.md to cover only the pipelined timing model with formulas grounded in PipelinedCheckpointTimingModel and a corrected 72 s / 8 s walkthrough. Ran /codex for a critical review and fixed all flagged issues (last-sub-slot-is-not-cooldown, event-emit timing, config env-var names, attestation-deadline nuance, insufficient-valid-txs handling, publisher preCheck semantics).

Changes

• sequencer-client: Replaced README.md with an architecture-first rewrite covering pipelining (build slot vs target slot, depth bound of 2, parent-invalidation discard), the per-slot job lifecycle, the publisher's Multicall3 flow, and the full config reference.
• sequencer-client (sequencer): Replaced src/sequencer/README.md with a pipelining-only timing model. Documents timeReservedAtEnd, maxNumberOfBlocks, per-state deadlines, proposer-vs-committee parallel timeline, and timing-variation handling.