Fix LAG member selection using packet data hash

[Devansh-567]

Problem

In McSimplePreLAG::replicate() (src/bm_sim/simple_pre_lag.cpp), the value used to pick a member port within a LAG was a hardcoded constant:

int lag_hash = 0xFF;  // TODO(unknown): get lag hash from metadata
...
port_count1 = (lag_hash % lag_entry.member_count) + 1;

Since lag_hash never varies, every packet replicated through a given LAG always resolves to the same member index, defeating the entire point of a LAG (spreading traffic across members).

Fix

• McSimplePre::McIn (include/bm/bm_sim/simple_pre.h) gains a new uint64_t hash{0} field. Callers that replicate over a LAG populate it with a hash derived from per-packet data; callers/targets that don't care about LAGs (e.g. l2_switch, which uses plain McSimplePre) are unaffected because it defaults to 0.
• McSimplePreLAG::replicate() now uses ingress_info.hash instead of the hardcoded 0xFF.
• simple_switch and psa_switch (the two targets that use McSimplePreLAG) now compute a hash of the packet's bytes with the existing bm::hash::xxh64 helper and pass it through: pre->replicate({mgid, lag_hash}). Hashing the packet's own bytes (rather than e.g. a monotonic counter) means packets from the same flow keep landing on the same LAG member, which is the behavior real LAG hashing is expected to have.

Testing

• Added McSimplePreLAG.LAGHashSelectsMember in tests/test_pre.cpp: creates a 4-member LAG and calls replicate() with several different hash values, asserting that the selected member follows the expected hash % member_count mapping and that all 4 members are reachable (this test would have failed before the fix, since every hash value used to resolve to the same member).
• Built and ran the full test suite locally (cmake + ctest, with WITH_THRIFT=ON -DWITH_NANOMSG=ON -DWITH_TARGETS=ON): 46/47 tests pass. The one failure (test_switch/SerializeState2) is unrelated to this
  change, it's an environment issue (missing Python thrift module for runtime_CLI.py in the sandbox used to test this), not a regression.
• Verified simple_switch and psa_switch (the two targets touched) build and their own test suites (test_packet_redirect, test_swap, test_queueing, test_recirc, test_parser_error, test_hash, test_internet_checksum) pass unchanged.
• Ran tools/cpplint.py on all changed files; confirmed zero new lint warnings versus the pre-change baseline (the warnings cpplint reports in psa_switch.cpp and simple_pre.h are pre-existing, unrelated to the lines touched here).

Files changed

• include/bm/bm_sim/simple_pre.h
• src/bm_sim/simple_pre_lag.cpp
• targets/simple_switch/simple_switch.cpp
• targets/psa_switch/psa_switch.cpp
• tests/test_pre.cpp