Fix Mbps pacing overflow on long replays

[AB-Coder96]

Standards checklist:

• The PR title is descriptive.
• The PR doesn't replicate another PR which is already open.
• The code is mine or it's from somewhere with an MIT-compatible license.
• The code is efficient, to the best of my ability, and does not waste computer resources.
• The code is stable and I have tested it myself, to the best of my abilities.
• The code passes sudo make test

Note: I noticed #982 after opening this PR. #982 appears to address the same overflow class and is broader because it touches both speed_mbpsrate and speed_packetrate.

This PR is a smaller, fix-only alternative focused on the speed_mbpsrate path, which I independently reproduced and validated. One difference is that this PR uses quotient/remainder arithmetic, which avoids multiplying the full bits_sent value by 1e9 while preserving the fractional remainder after division.

Changes:

Fixes #974.

This fixes an overflow in the speed_mbpsrate timing calculation for long low-rate replays.

Previously, next_tx_ns could be calculated as:

(bits_sent * 1000000000) / bps

For large captures, bits_sent * 1000000000 can overflow COUNTER before the division. When that happens, next_tx_ns can wrap to a much smaller value, which can make skip_length become very large. The send loop then skips sleep recalculation and can send the remaining packets too quickly near EOF.

This patch computes the same value using quotient/remainder arithmetic, avoiding multiplication of the full bits_sent value by 1e9.

Other comments:

Validated with a synthetic 22M-packet UDP pcap at -M 10.

Before fix:

• -T nano: final average around 10.11 Mbps, with a near-EOF final interval burst
• -T gtod: final average around 10.09 Mbps, with a near-EOF final interval burst

After fix:

• -T nano: expected 1865.60s, observed 1865.60s, final average 10.00 Mbps
• -T gtod: expected 1865.60s, observed 1865.62s, final average 9.99 Mbps

Test environment:

• Ubuntu 24.04.4 LTS under WSL2
• Kernel 6.6.87.2-microsoft-standard-WSL2
• dummy replay0 interface
• PF_PACKET send()

Also ran sudo make test successfully.

[AB-Coder96]

I noticed #982 after opening this PR. It appears to address the same overflow class and is broader because it touches both speed_mbpsrate and speed_packetrate.

One difference: this PR uses quotient/remainder arithmetic for the speed_mbpsrate calculation, so it avoids multiplying the full bits_sent value by 1e9 while preserving the remainder:

next_tx_ns = (bits_sent / bps) * 1000000000;
next_tx_ns += ((bits_sent % bps) * 1000000000) / bps;

#982 instead converts to nanoseconds after an earlier division, which also avoids the reproduced overflow but may lose some sub-millisecond precision due to integer truncation.

I’m happy to close this if maintainers prefer #982, or adjust this PR if the quotient/remainder approach is useful.