[lesnik512]

Closes #356.
Closes #357.

Three fixes to MemoryLoggerFactory.__call__, each test-first:

• One handler per logger (Each structlog proxy adds another MemoryHandler to the same stdlib logger, duplicating lines #356). structlog calls the factory once per get_logger() proxy, and every call attached another MemoryHandler to the same stdlib logger, so lines repeated once per proxy created so far. The factory now keeps a dict of the loggers it configured and returns a known logger unchanged.
• Caller's name for unnamed loggers (MemoryLoggerFactory names unnamed loggers after itself and resets a level set beforehand #357). The factory passes its own module to LoggerFactory(ignore_frame_names=...), so structlog.get_logger() is named after the caller's module instead of lite_bootstrap.instruments.logging_factory. The parameter is older than the structlog 21.3 floor.
• Preset levels kept (MemoryLoggerFactory names unnamed loggers after itself and resets a level set beforehand #357). logging_log_level is set only on a logger whose level is still NOTSET. close_handlers() resets the levels it set back to NOTSET. Without that, a second bootstrap with a different logging_log_level would keep the first run's level, and a separate test covers this.

Release-note line: a stdlib level a service set on a logger before bootstrap now takes effect for structlog calls on that logger. Before, it was silently overwritten with logging_log_level, so such a service may see more (or fewer) lines on that logger after upgrading.

Verified: the four new tests failed before the change, the full suite passes at 100% coverage, just lint is clean, and both reproductions from the issues now print one line per call, the caller's module name, and the preset DEBUG line.