Add Timestamp_Secs to V3 report + timestamp-based windowing (#581)

[marziehlenjaniMeta]

Summary:

Two related fixes to ensure perfpub selects the correct time window for
perf-collector-timeseries CSVs across all platforms.

1. V3 report generator: add missing Timestamp_Secs column

The Neoverse V3 report generator (generate_arm_neoversev3_perf_report.py)
defines a timestamp() function that emits a Timestamp_Secs column, but
never calls it in the metrics list. This column is already emitted by the
V2/Grace and AMD report generators. Without it, perfpub cannot do
timestamp-based row selection for V3 CSVs and must fall back to row-count
arithmetic.

Note: the index column in perf-collector-timeseries CSVs is a raw
DataFrame row number (jumps by 67-94 per row due to the multi-event-per-
timestamp structure of perf stat output), NOT a timestamp. Each row is
actually ~5 seconds apart. The Timestamp_Secs column provides the real
timestamp.

2. perfpub: use timestamp-based windowing for --last-secs/--skip-last-secs

When --last-secs and --skip-last-secs are specified, get_start_end_index()
previously used row-count arithmetic (len(df) - ceil(last_secs / interval)),
which depends on all CSVs having the same collection interval. This is fragile:
if different CSVs start collection at different times or have different
intervals, the row-count approach selects different absolute time windows.

Now, for CSVs with relative_secs or epoch_secs timestamp columns,
perfpub computes the target time window from the last timestamp
(end = last_ts - skip_last_secs, start = end - last_secs) and finds
the closest matching rows. This reuses the same idxmin() pattern already
used for the breakdown.csv path, ensuring all CSVs cover the same absolute
time window.

The row-count arithmetic is preserved as a fallback for time_of_day
CSVs (mpstat, memstat, etc.) and for old data lacking timestamp columns.

Differential Revision: D101064874

[meta-codesync]

@marziehlenjaniMeta has exported this pull request. If you are a Meta employee, you can view the originating Diff in D101064874.