perf: Skip BigQuery schema diff for unchanged payloads

[ruslandoga]

NOT YET READY FOR REVIEW

---

This PR:

• Adds SchemaBuilder.build_table_schema_with_change/2 so callers can reuse the existing schema and know whether a payload changed it.
• Updates BigQuery schema planning to use that change flag instead of rebuilding and then doing a full schema equality diff for unchanged payloads.
• Preserves build_table_schema/2 behavior and adds focused tests for changed and unchanged cases.
• Persists the branch profiler benchmark snapshot showing the reduced noop work.

[ruslandoga]

$ MIX_ENV=test LABEL="schema-builder-change-flag-vs-baseline" MACHINE="m2-local" mix run test/profiling/bq_schema_update_internals_bench.exs

##### With input edge log #####
Name             ips        average  deviation         median         99th %
update      105.52 K        9.48 μs   ±451.54%           9 μs       18.33 μs
noop         81.61 K       12.25 μs   ±273.77%       11.67 μs       18.04 μs

Memory usage statistics:
update        17.01 KB
noop          11.51 KB

Reduction count statistics:
update          1.85 K
noop            1.12 K

##### With input lists #####
Name             ips        average  deviation         median         99th %
noop        280.31 K        3.57 μs   ±451.53%        3.29 μs        6.71 μs
update      263.12 K        3.80 μs   ±442.34%        3.54 μs        5.33 μs

Memory usage statistics:
noop           4.61 KB
update         7.04 KB

Reduction count statistics:
noop               464
update             824

##### With input scalars #####
Name             ips        average  deviation         median         99th %
noop        208.36 K        4.80 μs   ±259.13%        4.50 μs        8.63 μs
update      206.18 K        4.85 μs   ±272.51%        4.50 μs       12.67 μs

Memory usage statistics:
noop           5.45 KB
update         8.23 KB

Reduction count statistics:
noop               532
update             939

Delta vs 14d2094d0 (2026-06-02T13:02:23.013962Z):
  noop / edge log
    ips:  34.2K ips → 81.6K ips  (+138.8%)
    wall: 26.58 → 11.67 µs  (-56.1%)
    mem:  35.7 KB → 11.5 KB  (-67.8%)
    reds: 2.9K → 1.1K  (-61.7%)
  noop / lists
    ips:  120.6K ips → 280.3K ips  (+132.4%)
    wall: 7.42 → 3.29 µs  (-55.7%)
    mem:  13.4 KB → 4.6 KB  (-65.5%)
    reds: 1.3K → 464  (-65.1%)
  noop / scalars
    ips:  83.2K ips → 208.4K ips  (+150.4%)
    wall: 10.54 → 4.5 µs  (-57.3%)
    mem:  17.2 KB → 5.4 KB  (-68.4%)
    reds: 1.5K → 532  (-65.4%)
  update / edge log
    ips:  29.9K ips → 105.5K ips  (+252.8%)
    wall: 32.08 → 9.0 µs  (-71.9%)
    mem:  40.7 KB → 17.0 KB  (-58.2%)
    reds: 3.2K → 1.8K  (-42.4%)
  update / lists
    ips:  109.1K ips → 263.1K ips  (+141.2%)
    wall: 8.17 → 3.54 µs  (-56.7%)
    mem:  14.5 KB → 7.0 KB  (-51.5%)
    reds: 1.4K → 824  (-42.3%)
  update / scalars
    ips:  75.0K ips → 206.2K ips  (+174.8%)
    wall: 11.83 → 4.5 µs  (-62.0%)
    mem:  18.5 KB → 8.2 KB  (-55.4%)
    reds: 1.6K → 939  (-40.7%)