feat: add streaming mode for large files (#58)

Adds --stream flag for element-by-element processing of JSON arrays,
JSONL, and CSV input. In streaming mode, each element is parsed,
optionally transformed via mappings, and written to the output format
immediately rather than buffering the entire dataset in memory.

Supported streaming pipelines:
- JSONL → JSONL/JSON/CSV (true line-by-line streaming)
- CSV → JSONL/JSON/CSV (true row-by-row streaming)
- JSON array → JSONL/JSON/CSV (parse + stream output)

Features:
- StreamWriter abstraction for incremental output
- Mapping support (per-element transforms via -e/-m)
- CSV delimiter support in streaming mode
- Graceful fallback: --stream is silently ignored for unsupported
  format combinations (e.g. YAML→JSON)
- Periodic flushing via BufWriter

Includes 24 unit tests and 7 integration tests covering all streaming
combinations, edge cases, and mapping integration.

Fixes #29

Author: alvinreal

src/cli.rs

@@ -150,6 +150,10 @@ pub struct Cli {
     /// Treat YAML input as multi-document (return array of documents)
     #[arg(long = "yaml-multi")]
     pub yaml_multi: bool,
+
+    /// Enable streaming mode for large files (processes elements one at a time)
+    #[arg(long = "stream")]
+    pub stream: bool,
 }
 
 impl Cli {
@@ -419,6 +423,11 @@ pub fn run(cli: &Cli) -> crate::error::Result<()> {
     let in_fmt = cli.resolve_input_format()?;
     let out_fmt = cli.resolve_output_format()?;
 
+    // Streaming mode: process elements one at a time for large files
+    if cli.stream && crate::streaming::can_stream(in_fmt, out_fmt) {
+        return crate::streaming::run_streaming(cli, in_fmt, out_fmt, mapping_program.as_ref());
+    }
+
     let input_data = read_input(cli)?;
     let value = parse_input_with_cli(&input_data, in_fmt, Some(cli))?;
 

src/lib.rs

@@ -2,4 +2,5 @@ pub mod cli;
 pub mod error;
 pub mod formats;
 pub mod mapping;
+pub mod streaming;
 pub mod value;