Add inline/literal input format for demos and testing

[rafafrdz]

Problem

Today input: requires format + path pointing to a physical file (csv/parquet/json). This makes it impossible to:

• Run a self-contained demo against a remote worker (e.g. Fly) that has no data on disk
• Write spec-level examples that execute without external fixtures
• Quickly prototype transformations without provisioning files

Proposal

Add an inline (a.k.a. literal / values) input format that embeds the rows directly in the YAML:

input:
  name: sales
  format: inline
  schema:
    - { name: order_id, type: int }
    - { name: product,  type: string }
    - { name: quantity, type: int }
    - { name: price,    type: double }
  rows:
    - [1, "Book",   2, 15.0]
    - [2, "Laptop", 1, 900.0]
    - [3, "Pen",    10, 1.5]

Alternative row syntax (object form) also acceptable:

rows:
  - { order_id: 1, product: "Book", quantity: 2, price: 15.0 }

Scope

• Spec: define inline format, required schema, supported primitive types, row validation rules
• Backends: DataFusion (MemTable), Polars (DataFrame::new), Spark (createDataFrame)
• Size guardrail: recommend cap (e.g. 1k rows) — this is for demos/tests, not production ingestion

Motivation

Unblocks end-to-end demos against deployed workers without needing volumes, uploads, or bundled fixtures.