Add ibis/xorq analysis backend - #518

- Fix IbisAnalysisPipeline: remove premature xorq guard (only require ibis),
  add schema pre-seeding for dtype, filter None from expression results,
  add histogram_query_fns support
- Add ibis_stats_v2.py with IbisTypingStats, IbisBaseSummaryStats,
  IbisNumericStats, IbisComputedSummaryStats, IbisHistogramStats
- Add xorq optional dependency group to pyproject.toml
- Add 23 tests using ibis.memtable (no xorq needed for local tests)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Replaces the existing v1-style class tests with tests targeting the
yet-to-exist IbisStatPipeline + @stat-based ibis_stats_v2 module.

Adds coverage that was missing in v1:
  - Live histogram (numeric + categorical) — was dead code
  - Structured error capture: a bad @stat surfaces as a StatError,
    not a silent stat dropout
  - DAG validation at pipeline construction
  - Constant-column histogram edge case

Will fail collection until ibis_stat_pipeline.py exists. Implementation
in the next commit.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Replaces the v1-style IbisAnalysis class hierarchy with @stat-decorated
functions executed through a new IbisStatPipeline that mirrors the
typed-DAG semantics of StatPipeline.

Two-phase execution:
  1. Batch aggregate — every @stat with an IbisColumn parameter is
     folded into a single table.aggregate() query and executed once.
  2. Per-column post-batch — computed stats and IbisTable-param stats
     (histograms) run through the standard typed-DAG executor.

Resolves the issues called out in the adversarial review:

  - Histogram is live (was dead code in v1: IbisHistogramStats was
    defined but never added to IBIS_ANALYSIS). Numeric and categorical
    paths now have test coverage that asserts populated histograms.

  - Structured error capture replaces the three `except Exception:
    continue` blocks. Failures inside an @stat or in the batch query
    surface as StatError entries; downstream computed stats still see
    UpstreamError propagation. Tests assert that an intentionally bad
    @stat shows up in the returned errors list.

  - DAG ordering is implicit. requires_summary is no longer a comment —
    parameter names ARE the dependency declaration, validated at
    construction via build_typed_dag.

  - Naming matches the v2 pattern: filename stays ibis_stats_v2 but the
    contents now match the @stat / TypedDict style of pd_stats_v2.

  - Empty-table length=0 fix: the previous `COUNT(c) + SUM(...IS NULL)`
    expression returned NULL on an empty table because SUM-over-empty
    is NULL in SQL. Coalesced to 0.

Adds IbisColumn / IbisTable marker types to stat_func.RAW_MARKER_TYPES
and a small IbisTable injection branch in _execute_stat_func so the
v2 pipeline can pass the table through to histogram-style stats.

Deletes the v1 IbisAnalysisPipeline / IbisAnalysis (no remaining
consumers in buckaroo/ or tests/).

25/25 ibis tests pass; full unit suite passes (modulo the pre-existing
unrelated MCP uvx flake on main).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Asserts the DfStats-shaped wrapper that lets DataFlow / BuckarooWidget
consume an ibis.Table: .sdf, .errs (v1 ErrDict shape), the
verify_analysis_objects classmethod, and .ap.ordered_a_objs.

Will fail collection until IbisDfStatsV2 is added in df_stats_v2.py.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Makes the ibis pipeline reachable from the rest of buckaroo:

  - Adds IbisDfStatsV2 in df_stats_v2.py mirroring DfStatsV2 /
    PlDfStatsV2 (.sdf, .errs, .ap.ordered_a_objs,
    verify_analysis_objects). Any DataFlow consumer that takes an
    arbitrary DFStatsClass now also accepts the ibis path.

  - Adds process_table_v1_compat to IbisStatPipeline so the wrapper
    returns the same {(col, stat): (Exception, kls)} ErrDict shape
    that AnalysisPipeline produced.

  - Adds ordered_a_objs property to IbisStatPipeline so the
    add_analysis path matches StatPipeline.

ibis is still imported lazily inside IbisDfStatsV2 — the rest of the
df_stats_v2 module stays ibis-free.

End-to-end test asserts the round-trip (table -> .sdf with length=5
and mean=3.0) and that v1 errs shape matches the existing convention.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>