OpenAPI 3.1 normalizer (#1)

[keysersoft]

Problem

FastAPI ≥ 0.100, recent NestJS, spring-doc 3.x and Hono all emit OpenAPI 3.1 natively. Our pipeline depends on swagger-parser 10.x which hard-rejects any `openapi` value other than 3.0.0–3.0.3 — including from `dereference()`. Customers were getting `Unsupported OpenAPI version: 3.1.0` and had to patch their bridges to emit 3.0 by hand (PR #5 in the customer's repo did this).

Fix

A small normalizer that rewrites the 3.1-only constructs we care about into their 3.0 equivalents in place, before swagger-parser sees the document, and relabels `openapi` to `3.0.3` so the parser proceeds.

Rewrites covered:

3.1 input	3.0 output
`type: [X, "null"]`	`type: X` + `nullable: true`
`anyOf: [X, {type:"null"}]`	unwrap `X` + `nullable: true`
`oneOf: [X, {type:"null"}]`	same
`const: "x"`	`enum: ["x"]`
`examples: [...]` (array)	`example: examples[0]`
`exclusiveMinimum: `	`minimum: ` + `exclusiveMinimum: true`
`exclusiveMaximum: `	`maximum: ` + `exclusiveMaximum: true`

The `components.examples` map (object, not array) is preserved.

Also forwards `nullable` and `example` through `paramToJsonSchema` / `flattenSchema` so the normalized fields actually survive into the generated tool schemas. Without this, the normalizer's work was being silently dropped when the parser flattened parameters.

Out of scope

• Webhooks, jsonSchemaDialect, info.summary — survive untouched, not used by tool extraction.
• Full replacement of swagger-parser (the "stradone giusto" with `@redocly/openapi-core`) — left for a future PR.

Test plan

• 16 unit tests for the normalizer (`openapi-3.1-normalizer.spec.ts`).
• 3 E2E tests in `openapi.parser.spec.ts` covering a FastAPI-style 3.1 spec end-to-end.
• Full backend test suite: 620 passed, 0 regressions.