feat: Native Broadcast nested loop join support

[coderfender]

Which issue does this PR close?

Closes #198

Wires Spark's BroadcastNestedLoopJoinExec to DataFusion's NestedLoopJoinExec.

Rationale for this change

Native support for BroadcastNestedLoopJoin to improve native performance.

What changes are included in this PR?

How are these changes tested?

1. Unit tests in CometJoinSuite
2. Setup BNLJ benches

[coderfender]

Bunch TPCDS queries seem to be using BNLJ warranting golden file regen

[coderfender]

Benchmarks (local M5 pro) (Edited after running benchmarks with JIT disabled) :

Running benchmark: range join (BETWEEN)
  Running case: Spark
  Stopped after 23 iterations, 2043 ms
  Running case: Comet
  Stopped after 33 iterations, 2051 ms

OpenJDK 64-Bit Server VM 17.0.16+8-LTS on Mac OS X 26.3.2
Apple M5 Pro
range join (BETWEEN):                     Best Time(ms)   Avg Time(ms)   Stdev(ms)    Rate(M/s)   Per Row(ns)   Relative
------------------------------------------------------------------------------------------------------------------------
Spark                                                73             89          10         14.4          69.7       1.0X
Comet                                                49             62          11         21.4          46.8       1.5X

Running benchmark: inequality join (>)
  Running case: Spark
  Stopped after 21 iterations, 2038 ms
  Running case: Comet
  Stopped after 34 iterations, 2028 ms

OpenJDK 64-Bit Server VM 17.0.16+8-LTS on Mac OS X 26.3.2
Apple M5 Pro
inequality join (>):                      Best Time(ms)   Avg Time(ms)   Stdev(ms)    Rate(M/s)   Per Row(ns)   Relative
------------------------------------------------------------------------------------------------------------------------
Spark                                                78             97          19         13.4          74.4       1.0X
Comet                                                48             60           6         21.8          46.0       1.6X

Running benchmark: left outer non-equi
  Running case: Spark
  Stopped after 27 iterations, 2028 ms
  Running case: Comet
  Stopped after 35 iterations, 2028 ms

OpenJDK 64-Bit Server VM 17.0.16+8-LTS on Mac OS X 26.3.2
Apple M5 Pro
left outer non-equi:                      Best Time(ms)   Avg Time(ms)   Stdev(ms)    Rate(M/s)   Per Row(ns)   Relative
------------------------------------------------------------------------------------------------------------------------
Spark                                                63             75          11         16.6          60.1       1.0X
Comet                                                42             58          16         25.2          39.7       1.5X

[coderfender]

TODO : Build vs Probe swap to DF's BNLJ op

[coderfender]

Investigating test failures

[mbutrovich]

First round of feedback. Thanks @coderfender! This is looking really good!

[mbutrovich]

The support matrix here lines up exactly with Spark's BroadcastNestedLoopJoinExec.outputPartitioning supported set, and the reasoning in broadcastBuildReplicationReason is spot on. Every case returns either Compatible(None) or Unsupported(...), never Incompatible. Since isOperatorEnabled only consults the allow-incompat config on the Incompatible branch, the getOperatorAllowIncompatConfigKey("BroadcastNestedLoopJoinExec") line in the CometJoinSuite override is dead. Is that intentional, or leftover?

[coderfender]

Agreed. I was initially gating the join behind the config in the test suite to figure out the support matrix through testing. I have since then removed the now redundant config