fix(logger): dedicated console logger for the inference status table

[ff225]

Closes #66. Follow-up to #12 / PR #60.

Problem

PR #60 replaced the inference-table print() calls in request_handler.py with logger.info(). Every table row then went through the logger formatter (src/server/logger/log.py), which:

• prefixes each row with a ~70-char asctime - level - file:func:line string, breaking the column alignment, and
• interleaves the table with the verbose diagnostic stream (network speed, model registration, delay sim, profiler), burying the "which layer / what timings" signal.

Fix

Add a dedicated table_logger (server.table) decoupled from the diagnostic logger:

• plain %(message)s formatter on stdout (matches the pre-PR-60 print() target; default StreamHandler is stderr),
• propagate=False so it never double-emits through root,
• fixed at INFO so the table stays visible even when verbose: false silences the noisy INFO/DEBUG stream.

_print_inference_table_header / _print_inference_table_row now use table_logger.

Testing

Verified manually before commit:

• table rows emit prefix-free and aligned;
• table stream is isolated from the diagnostic logger (no leakage either way);
• table remains visible after the non-verbose config raises logger/root to CRITICAL.

tests/unit/test_no_print_statements.py (added in PR #60) still passes.