feat: v0.3.0 — asymmetric K/V bits, layer-adaptive precision, deprecate QJL

overhaul: rewrite README hook, clean repo, fix positioning

docs: add active plan for v0.2.0 launch and next steps

feat: 4-bit nibble packing — halves index storage, 1 GB saved at 4K context

marketing: add strategy, blog post, reddit posts, portfolio blog plan

docs: fix last stale refs — PyPI 0.1.0→0.2.0, roadmap status update

docs: update all references for v0.2.0 compressed index storage

feat: compressed index storage — real KV cache compression (v0.2.0)

docs: accuracy sweep — fix stale numbers, update references, add research index

bench: add cross-architecture + long-context data (42 total data points)

bench: RTX 4080 benchmark data — 7B + 3B Qwen, context sweep

docs: add ExecPlan workflow system to guides

docs: add archived plans (moved from FlockRun)

docs: add KV cache compression research (moved from FlockRun)

docs: full doc system — architecture, reference, workflow, codebase map