Reconcile chunked-64 packing (PR #225) with #234 single-run packing #240

PR #225 lands chunked-64 packed int vectors (PACKED_CHUNK=64, frozen chunks in an rpds spine + tail), replacing the MAX_PACKED_LEN=1024 single-run packing from #234 (including re-pinned cap tests).

Rationale for landing chunked first: O(1) persistent conj/assoc/pop at any size with ~8B/elem density, vs cap-then-generic-fallback; verified by the PR's 23/23 head-to-head wins vs JVM Clojure and the full suite at 100% line coverage.

Follow-up work:

1. Run Cap packed Int runs at 1024 to fix #227 case regressions #234's quadratic-chain workloads plus the Attack RS slow legs: pooled recur, string fast paths, shape cadence (#235) #238 proxies (boyer/suffix/segment) against chunked-64 packing and compare with the Cap packed Int runs at 1024 to fix #227 case regressions #234 numbers.
2. Decide the single long-term design: keep chunked, adopt Cap packed Int runs at 1024 to fix #227 case regressions #234's single-run, or unify (e.g. chunked storage with Cap packed Int runs at 1024 to fix #227 case regressions #234-style promotion policy).
3. Unify or explicitly retire the losing variant's tests/docs.

Raised in PR #225 round-1 review (michael-h-hu).