Fix poor performance of table reading when many record batches are involved

[quinnj]

Fixes #528. Alternative to #568

cc: @KristofferC

[KristofferC]

Before:

❯ julia --project mwe.jl
  0.140257 seconds (2.56 M allocations: 130.583 MiB, 16.17% gc time

❯ julia --project --threads=auto mwe.jl
 32.504022 seconds (2.58 M allocations: 131.676 MiB, 0.04% gc time, 13046169 lock conflicts)

After:

❯ julia --project mwe.jl
  0.137567 seconds (2.71 M allocations: 142.244 MiB, 14.55% gc time)

❯ julia --project --threads=auto mwe.jl
  0.102691 seconds (2.70 M allocations: 141.407 MiB, 8.58% gc time, 65 lock conflicts)

🚀