If $L_1,L_2$ are two disjoint context-free languages which are not inherently ambiguous, then $L_1 \cup L_2$ is also a context-free language which is not inherently ambiguous. The reason is simple: starting with context-free grammars for $L_1,L_2$ with start symbols $S_1,S_2$ (respectively) and otherwise disjoint non-terminals, you can construct an unambiguous grammar for $L_1 \cup L_2$ by adding a new start symbol $S$ and the productions $S\to S_1|S_2$. This takes care of 1) and 3), once you construct unambiguous grammars for the constituent languages.
As Palec mentions, 4) is regular, and so admits a regular grammar which corresponds to a DFA accepting it. This grammar has as non-terminals the states of the DFA, and productions $A \to aB$ if the DFA moves from $A$ to $B$ upon reading $a$. For each final state $F$ we add a production $F\to\epsilon$, and the start symbol is the start state. This description should make it clear that the grammar is unambiguous.
Intuitively, 2) could be inherently ambiguous since words of the form $a^n b^m c^p$ with $n \neq m \neq p$ could have two different parse trees, corresponding to the two constituent languages. Proving this formally could be tedious. The only method I know consists of coming up with two words $x,y$ that can be pumped to the same word $w$ in such a way that makes it clear that the corresponding parse trees are different. If you want to pursue this further, a good first step will be to read some examples of this method.
