fix(sequential-thinking): preserve nextThoughtNeeded in required schema fields

[AbhiPra24]

Description

Fixes #4651.

In #3533, boolean parameters in src/sequentialthinking/index.ts were switched to use z.preprocess() to support string booleans from LLMs. However, z.preprocess() wraps the underlying schema into a pipeline accepting unknown, causing Zod's JSON schema generation (z.toJSONSchema) to omit nextThoughtNeeded from the required properties list.

This caused MCP clients constructing calls based strictly on the declared required tool schema to omit nextThoughtNeeded and receive validation error -32602 at runtime.

This PR fixes the issue by replacing z.preprocess() with z.union([z.boolean(), z.string().transform(...)]). This preserves case-insensitive string-to-boolean coercion while correctly emitting nextThoughtNeeded in the tool's inputSchema.required array.

Changes

• Update coercedBoolean definition in src/sequentialthinking/index.ts to z.union.
• Add unit tests in src/sequentialthinking/__tests__/schema.test.ts verifying input schema generation, required fields, and boolean/string parsing.

Verification

• Ran npm test and npm run build in src/sequentialthinking. All 18 tests pass with 100% coverage on schema logic.

[olaservo]

The same fix landed via #4652 (d6402efe): coercedBoolean is now a union + transform instead of z.preprocess, so toJSONSchema keeps nextThoughtNeeded in required. The merged regression test asserts against the built server's advertised schema over stdio, which also covers what the test here checks. Closing as superseded — the union approach you proposed matches what was merged.