[None][perf] Skip transceiver tp_allgather when no sessions ever opened

[Shixiaowei02]

@coderabbitai summary

Description

V4-Pro 8k/1k MTP3 disagg (ctx1 TP=4 + gen4 TP=8), concurrency=4/server (total 16), 128 prompts × 8 rounds, gb300:

Metric	Before	After	Δ
output_throughput (tok/s)	600.1	626.5	+4.41%
user_throughput (tok/s)	45.87	47.38	+3.28%
mean_tpot_ms	6.373	6.126	−3.87%
mean_ttft_ms	16491	16300	−1.16%
mean_itl_ms	1434.9	1416.4	−1.29%
mean_e2el_ms	22354	21932	−1.89%
avg_decoded_tokens_per_iter	2.654	2.676	+0.83%
duration (s)	196.2	188.0	−4.22%

GPU compute is unchanged . The full 4.4% e2e throughput win comes from eliminating one tp_allgather per decode iter on the GEN worker, which had previously synced all 8 TP ranks every iter against an empty consensus list.

Test Coverage

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[Shixiaowei02]

/bot run --add-multi-gpu-test --disable-fail-fast

[tensorrt-cicd]

PR_Github #47973 [ run ] triggered by Bot. Commit: 70162aa Link to invocation

[tensorrt-cicd]

PR_Github #47973 [ run ] completed with state SUCCESS. Commit: 70162aa
/LLM/main/L0_MergeRequest_PR pipeline #37813 completed with status: 'SUCCESS'

CI Report

Link to invocation