Add delta report for benchmarks

[gabotechs]

Adds a delta report that compares the last run benchmarks with the current ones, something like this:

==== Comparison with the previous benchmark from 2025-08-12 14:38:33 UTC ====
 Query 1: prev=  74 ms, new=  70 ms, diff=1.06 faster ✔
 Query 2: prev=  25 ms, new=  25 ms, diff=1.00 slower ✖
 Query 3: prev=  37 ms, new=  36 ms, diff=1.03 faster ✔
 Query 4: prev=  25 ms, new=  28 ms, diff=1.12 slower ✖
 Query 5: prev=  46 ms, new=  54 ms, diff=1.17 slower ✖
 Query 6: prev=  22 ms, new=  21 ms, diff=1.05 faster ✔
 Query 7: prev=  64 ms, new=  61 ms, diff=1.05 faster ✔
 Query 8: prev=  47 ms, new=  43 ms, diff=1.09 faster ✔
 Query 9: prev=  63 ms, new=  53 ms, diff=1.19 faster ✔
Query 10: prev=  66 ms, new=  53 ms, diff=1.25 faster ✅
Query 11: prev=  34 ms, new=  21 ms, diff=1.62 faster ✅
Query 12: prev=  58 ms, new=  41 ms, diff=1.41 faster ✅
Query 13: prev=  34 ms, new=  32 ms, diff=1.06 faster ✔
Query 14: prev=  35 ms, new=  30 ms, diff=1.17 faster ✔
Query 15: prev=  53 ms, new=  48 ms, diff=1.10 faster ✔
Query 16: prev=  23 ms, new=  19 ms, diff=1.21 faster ✅
Query 17: prev=  77 ms, new=  70 ms, diff=1.10 faster ✔
Query 18: prev= 103 ms, new=  92 ms, diff=1.12 faster ✔
Query 19: prev=  58 ms, new=  51 ms, diff=1.14 faster ✔
Query 20: prev=  52 ms, new=  41 ms, diff=1.27 faster ✅
Query 21: prev= 112 ms, new=  89 ms, diff=1.26 faster ✅
Query 22: prev=  22 ms, new=  17 ms, diff=1.29 faster ✅

The first time the benchmarks are run, no report will be printed to stdout, the second time, a report will be produced comparing it with the first run, the third time will be compared to the second one, etc...

This can be tested running this command twice:

cargo run -p datafusion-distributed-benchmarks --release -- tpch --path benchmarks/data/tpch_sf1 

[robtandy]

Approved, but it will have conflicts for port number if #89 is merged first. I suggest we do it in that order to accommodate our staggered working hours.