db/datastruct/btindex: EF-encode .bt pivot offsets

[sudeepdino008]

Stacks on #21875. After that PR, the .bt pivot cache's only per-node heap cost is the offset array — a flat []uint64 (8 B/node). Those offsets are strictly increasing, so this stores them Elias-Fano-encoded (~7 bits/node) as the only representation: no flat array, no toggle.

The EF is built at open, directly from the existing nodes section — no on-disk format change, and both layouts are covered (v2 footer and v0 legacy list).

Heap (open pivot cache, commitment, M=256)

file	pivots (N/M)	flat []uint64 (before)	EF (this PR)
141M keys	543K	4.35 MB	0.46 MB (9.5×)
1M keys	4.2K	34 KB	5 KB

≈ 6.8 bits/pivot. On a mainnet node the aggregate pivot heap (storage-dominated, ~340 MB after #21875) drops to ~36 MB.

Latency / major-faults (cold = evicted page cache, 15k sampled keys)

file	variant	cold µs/op	cold faults/op	warm µs/op
141M keys	flat	140	1.57	10
141M keys	EF	143	1.57	8
1M keys	flat	180	2.54	6
1M keys	EF	181	2.54	7

Identical fault count, latency within run-to-run noise: the extra EF.Get calls in the pivot search are pure CPU and don't register against the ~80 µs/major-fault cost.

Tests

Test_BtreeIndex_NodeOfftEF_V0_V2 opens both v0 and v2 files and checks every key resolves to its true .kv value through the EF-decoded pivots. decodeNodes / decodeListNodesV0 unit tests updated for the EF return. Full btindex suite passes.

[AskAlexSharov]

i guess we can add it to .bt (to speedup startup). but i would do it after #20180 (because maybe it will work well - but may require "another datastructure" than EF of offsets)

[AskAlexSharov]

┌──────────────────────┬───────────────────────┬────────────────┬────────┐
  │      Benchmark       │ sudeep/bt-nodeofft-ef │ alex/bt_ram_36 │   Δ    │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTreeSeek          │ 670.6 ns/op           │ 566.0 ns/op    │ −15.6% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTreeGet           │ 539.9 ns/op           │ 424.5 ns/op    │ −21.4% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N1M_M256  │ 167.3 ns/op           │ 47.5 ns/op     │ −71.6% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N1M_M64   │ 175.9 ns/op           │ 50.8 ns/op     │ −71.1% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N10M_M256 │ 202.8 ns/op           │ 59.4 ns/op     │ −70.7% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M256     │ 474.2 ns/op           │ 349.2 ns/op    │ −26.4% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M128     │ 474.0 ns/op           │ 352.0 ns/op    │ −25.7% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M64      │ 475.6 ns/op           │ 350.3 ns/op    │ −26.3% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M32      │ 476.5 ns/op           │ 347.8 ns/op    │ −27.0% │
  └──────────────────────┴───────────────────────┴────────────────┴────────┘

[sudeepdino008]

┌──────────────────────┬───────────────────────┬────────────────┬────────┐
  │      Benchmark       │ sudeep/bt-nodeofft-ef │ alex/bt_ram_36 │   Δ    │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTreeSeek          │ 670.6 ns/op           │ 566.0 ns/op    │ −15.6% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTreeGet           │ 539.9 ns/op           │ 424.5 ns/op    │ −21.4% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N1M_M256  │ 167.3 ns/op           │ 47.5 ns/op     │ −71.6% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N1M_M64   │ 175.9 ns/op           │ 50.8 ns/op     │ −71.1% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BpsTree_bs/N10M_M256 │ 202.8 ns/op           │ 59.4 ns/op     │ −70.7% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M256     │ 474.2 ns/op           │ 349.2 ns/op    │ −26.4% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M128     │ 474.0 ns/op           │ 352.0 ns/op    │ −25.7% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M64      │ 475.6 ns/op           │ 350.3 ns/op    │ −26.3% │
  ├──────────────────────┼───────────────────────┼────────────────┼────────┤
  │ BtIndex_Get/M32      │ 476.5 ns/op           │ 347.8 ns/op    │ −27.0% │
  └──────────────────────┴───────────────────────┴────────────────┴────────┘

yeah, cpu will be more because of ef; things i cared about more -- not much effect on majfaults + heap reduction and fast startup.