This allows adding a new 'sample' submodule and enables invoking the
sampling profiler through 'python -m profile.sample', while retaining
backwards compatibility when using 'profile'.

Implement a statistical sampling profiler that can profile external
Python processes by PID. Uses the _remote_debugging module and converts
the results to pstats-compatible format for analysis.

This variant overrides how column headers are printed to avoid
conflating call counts with sample counts.

The SampledStats results are stored in the exact same format as Stats,
but since the results don't represent call counts but sample counts, the
column headers are different to account for this.

This ensure that using the pstats browser instantiates the right object
to handle the correct columns, add a factory function which can
instantiate the correct class. As the Stats class can only handle either
a filename or an object which provides the 'stats' attribute in a
pre-parsed format, this provides a StatsLoaderShim to avoid marshalling
the data twice (once to check the marker and once in the Stats module if
we were to pass the file name).

Implements collapsed stack trace format output for the sampling
profiler. This format represents complete call stacks as semicolon-
delimited strings with sample counts, making it compatible with
external flamegraph generation tools like flamegraph.pl.

The format uses filename:function:line notation and stores call trees
during sampling for efficient post-processing into the collapsed format.