Add `Start Tracing` / `Stop Tracing` / `Start Trace Chunk` / `Stop Trace Chunk` keywords for ad-hoc tracing

Use case

Today tracing is all-or-nothing: it can only be enabled for a whole context via the tracing argument of New Context, producing one trace for everything that context ever did. Users debugging a single flaky test case, or wanting a small shareable trace of just one scenario, cannot start tracing mid-test, cannot stop it early, cannot give the trace a human-readable title for the trace viewer, and cannot include source files. The only workaround is restructuring suites so that the interesting steps run in their own dedicated context — invasive and often impossible.

Playwright's Tracing API supports exactly this: start/stop on demand, plus chunks — multiple separate trace files recorded within one tracing session, each covering only the actions between startChunk and stopChunk.

Proposed keyword / arguments

Start Tracing    *, screenshots=True    snapshots=True    sources=False    title=None    name=None    live=False
Stop Tracing    path=None
Start Trace Chunk    *, title=None    name=None
Stop Trace Chunk    path=None

• Start Tracing (Setter) begins tracing on the current context. title names the trace in the trace viewer; sources=True embeds source files; live enables live-viewable traces; name prefixes intermediate trace files in the traces dir.
• Stop Tracing (Getter) stops tracing, saves the zip to path (default: a generated file under the library's output directory) and returns the file path.
• Start Trace Chunk / Stop Trace Chunk record only the actions in between into a separate trace file within the same tracing session, so one long-lived context can yield one small trace per scenario.

New Context
Start Tracing    title=Login flow    sources=True
Go To    ${LOGIN_URL}
Start Trace Chunk    title=Submit credentials
Fill Text    id=user    alice
Click    id=submit
${chunk}=    Stop Trace Chunk    path=${OUTPUT_DIR}/traces/login-submit.zip
${trace}=    Stop Tracing    path=${OUTPUT_DIR}/traces/login-full.zip

Playwright API

• tracing.start (options screenshots, snapshots, sources, title, name, live)
• tracing.stop
• tracing.startChunk / tracing.stopChunk

Implementation notes

• protobuf/playwright.proto: RPCs for start/stop/startChunk/stopChunk with the option fields.
• node/playwright-wrapper: calls on the current context's tracing object; must coexist with the existing New Context tracing= mechanism (guard against double-start with a clear error). Note the library already wraps every keyword in tracing.group internally — chunk boundaries slot into that unchanged.
• Python: keywords in Browser/keywords (natural fit: playwright_state.py or a new tracing module), docs, stub regen; atest verifying a valid trace zip is produced mid-context and that chunks contain only their own actions.

Backwards compatibility

Purely additive: four new keywords. The existing New Context tracing= shortcut keeps working unchanged; starting tracing twice is an explicit error rather than a behavior change.

Start with protobuf/playwright.proto and the calls in node/playwright-wrapper, then inspect the keyword patterns in Python's Browser/keywords. Run the atest coverage described in the issue, checking that a mid-context trace produces a valid zip and that chunks contain only their own actions. Done also requires the existing New Context tracing= behavior to remain intact and double-starting to report a clear error.