Expose SessionContext.add_analyzer_rule for Python-defined analyzer rules

## Background

`SessionContext::add_analyzer_rule` registers an `Arc` that runs during plan analysis (the pre-optimization pass that resolves names, applies type coercion, etc.). The Python bindings do not expose this method.

## Upstream signature

```rust
pub fn add_analyzer_rule(&self, analyzer_rule: Arc)
```

## User value

Lets users intercept and transform a logical plan immediately after analysis, before the main optimizer runs. Useful for custom semantic validation, injecting domain-specific defaults, or rewriting unresolved references against an external catalog.

## Why deferred

Blocked upstream, same root cause as #1574 ([add_optimizer_rule](https://github.com/apache/datafusion-python/issues/1574)): no FFI bridge exists for the logical `AnalyzerRule` trait, and a Python-only adapter can observe but not meaningfully transform `LogicalPlan` nodes because there are no Python constructors for the variants. Tracked here; revisit when upstream lands an `FFI_AnalyzerRule`.

Start by reviewing SessionContext::add_analyzer_rule, issue #1574, and the upstream work for an FFI_AnalyzerRule bridge. The task is ready when Python bindings can expose analyzer rules that meaningfully transform logical plans, including the required LogicalPlan support.

索引モデルが issue の本文から書いたものです。