fix(lambda): uppercase LOG_LEVEL in the shared lambda module

[zxkane]

Description

#5238 wrapped every LOG_LEVEL assignment with upper() so Powertools v2 stops silently discarding the value, but it missed modules/lambda/main.tf. That module builds the environment for four functions:

• modules/runners/job-retry
• modules/termination-watcher/deregister-retry
• modules/termination-watcher/notification
• modules/termination-watcher/termination

var.lambda.log_level is validated against lowercase names only (debug, info, warn, error), so the value reaching those four functions can never match Powertools' uppercase LogLevelThreshold keys — every level silently resolves to INFO. Asking for debug produces no debug output, and asking for warn or error produces more output than requested.

POWERTOOLS_LOGGER_LOG_EVENT still compares the raw lowercase variable, which remains correct, so it is left untouched.

Test Plan

Reproduced against the @aws-lambda-powertools/logger version this repository already depends on:

LOG_LEVEL	effective level
debug	INFO
DEBUG	DEBUG
warn	INFO
WARN	WARN

Confirmed on a live deployment before this change: the functions created through modules/lambda had LOG_LEVEL=info, while every function fixed by #5238 had LOG_LEVEL=INFO.

Checks run locally:

• terraform fmt -recursive -check=true -write=false
• terraform validate on modules/lambda, modules/runners/job-retry, modules/termination-watcher, modules/termination-watcher/notification and modules/termination-watcher/termination
• Confirmed no lowercase LOG_LEVEL assignment remains anywhere under modules/

Applying this to a running deployment is an in-place Lambda environment update — no resource replacement.

Related Issues

Follow-up to #5238.