Add measured decompression throughput to the array tree display

[joseph-isaacs]

Summary

Add a array tree display mode to see decompression throughput

Example, from the decompress benchmark's own trees:

== dict_fsst_varbin_string ==
root: vortex.dict(utf8, len=2048) nbytes=8.77 kB (100.00%)
  throughput: 63.27µs (100.00%) | in 138.52 MB/s | out 776.81 MB/s | 32.37 Mrow/s | self 61.91µs
  codes: vortex.primitive(u32, len=2048) nbytes=8.19 kB (93.46%)
    throughput: 67ns (0.11%) | in 122.27 GB/s | out 122.27 GB/s | 30.57 Grow/s | self 67ns
  values: vortex.fsst(utf8, len=20) nbytes=573 B (6.54%)
    throughput: 1.30µs (2.06%) | in 440.09 MB/s | out 368.66 MB/s | 15.36 Mrow/s | self 1.17µs

== for_bp_u64 ==
root: fastlanes.for(u32, len=2048) nbytes=2.05 kB (100.00%)
  throughput: 548ns (100.00%) | in 3.74 GB/s | out 14.95 GB/s | 3.74 Grow/s | fuses children (saves 170ns)
  encoded: fastlanes.bitpacked(u32, len=2048) nbytes=2.05 kB (100.00%)
    throughput: 718ns (131.02%) | in 2.85 GB/s | out 11.41 GB/s | 2.85 Grow/s | self 718ns

Root times track the decompress benchmark's medians within a few percent, so the breakdown can
be read against that benchmark directly:

tree	divan median	profile root
for_bp_u64	584.3 ns	548 ns
alp_for_bp_f64	4.676 µs	4.64 µs
dict_varbinview_string	2.327 µs	2.32 µs
runend_for_bp_u32	1.979 µs	1.91 µs
dict_fsst_varbin_string	64.88 µs	63.27 µs
dict_fsst_varbin_bp_string	43.91 µs	43.93 µs
datetime_for_bp	8.733 µs	8.69 µs