[abhinav]

Branch graph construction loads tracked state, but restack checks then
look up each branch again.
Pass the graph's existing branch records to
CheckLoadedBranchRestacked so listings avoid those repeated reads.
Keep graph loading behind Service.BranchGraph and share restack
reconciliation with CheckRestacked, including the recovered base hash
used to render a healed boundary.

On the macOS arm64 demo repository with 103 local branches, hyperfine
uses 3 warmups and 15 runs of log short and log long, each with
--all --json.
Compared with the parent branch-batching commit:

Listing	Parent mean +/- SD (s)	New mean +/- SD (s)
Short	1.043 +/- 0.127	0.590 +/- 0.157
Long	1.178 +/- 0.096	0.684 +/- 0.030

The collective effect of this stack is:

Benchmark 1: git-spice ls (10 runs)
                  mean ±     σ    min …   max
  Wall Time [s]  1.596 ± 0.388  1.269 … 2.396
  Memory [MiB]    24.0 ±   0.5   23.2 …  24.6

Benchmark 2: git-spice-new ls (10 runs)
                  mean ±     σ    min …   max
  Wall Time [s]  0.532 ± 0.064  0.488 … 0.699   -66.7%  (3.0x faster)
  Memory [MiB]    24.3 ±   0.2   23.9 …  24.6    +1.2%

[abhinav]

This change is part of the following stack:

• git: Add streaming batch object reads #1480
  • storage: Add batch reads to the backend contract #1481
    • spice: Batch branch state and head loading #1489
      • log: Reuse loaded branch state in listings #1490 ◀

_{Change managed by git-spice.}