Less confidently mark f-strings as empty when inferring truthiness (#20152)
When computing the boolean value of an f-string, we over-eagerly
interpreted some f-string interpolations as empty. In this PR we now
mark the truthiness of f-strings involving format specs, debug text, and
bytes literals as "unknown".
This will probably result in some false negatives, which may be further
refined (for example - there are probably many cases where
`is_not_empty_f_string` should be modified to return `true`), but for
now at least we should have fewer false positives.
Affected rules (may not be an exhaustive list):
- [unnecessary-empty-iterable-within-deque-call
(RUF037)](https://docs.astral.sh/ruff/rules/unnecessary-empty-iterable-within-deque-call/#unnecessary-empty-iterable-within-deque-call-ruf037)
- [falsy-dict-get-fallback
(RUF056)](https://docs.astral.sh/ruff/rules/falsy-dict-get-fallback/#falsy-dict-get-fallback-ruf056)
- [pytest-assert-always-false
(PT015)](https://docs.astral.sh/ruff/rules/pytest-assert-always-false/#pytest-assert-always-false-pt015)
- [expr-or-not-expr
(SIM221)](https://docs.astral.sh/ruff/rules/expr-or-not-expr/#expr-or-not-expr-sim221)
- [expr-or-true
(SIM222)](https://docs.astral.sh/ruff/rules/expr-or-true/#expr-or-true-sim222)
- [expr-and-false
(SIM223)](https://docs.astral.sh/ruff/rules/expr-and-false/#expr-and-false-sim223)
Closes #19935