Avoid exhausting file descriptors while collecting Linux processes

[fzlzjerry]

Fixes #766. Related: #743.

The second sampling pass retains every procfs::Process until the columns are populated. Each owns a directory descriptor, so a low open-file limit silently truncates the result.

Collect Linux samples lazily and populate all columns and parent maps for each process before advancing. This keeps the existing handle-based reads, without adding extra PID-based reopenings or changing resource limits. Other platforms keep their existing collectors.

The regression test builds a synthetic procfs tree with 96 processes and their threads, then runs the CLI with a child-only RLIMIT_NOFILE=32. It covers normal output, threads, tree output, kernel-thread filtering and command-line reads. Output must match the 1024-descriptor control. On the original code, the first case shows only 28 of 96 processes.

Validation:

• Rust 1.88.0 and 1.98.0, default and no-default features: all test binaries pass in private PID namespaces (25 tests with default features, 20 without, for each version).
• Rust 1.98.0: cargo clippy --locked --all-targets -- -D warnings, also with --no-default-features; formatting passes.
• A network-isolated, non-root container with 97 real processes: all 97 remain visible at limits 16, 32 and 1024; all 100 entries appear with threads or tree mode. TCP/UDP ports and working directories are preserved. Checked with both Rust versions.
• With 4096 synthetic processes, the original needs the higher limit to list them all and reaches 4100 observed open descriptors. The change lists all 4096 at a limit of 32, with an observed peak of 6 descriptors.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.